74 F.3d 1571
9th Cir.
1996

Lead Opinion

Opinion by Judge KLEINFELD.

KLEINFELD, Circuit Judge:

This is a state habeas corpus case. We apply the newly established standard for prejudice in state habeas constitutional error cases, and determine that on the facts of this ease, there was no prejudice requiring that the writ be issued.

I. Facts.

Two groups confronted each other behind a fried chicken restaurant. Thompson stabbed three men from the other group, killing one of them. He also held a knife as he spoke with a woman in a car and slashed one of her tires.

He was charged with first degree murder for the killing, attempted murder for the two stabbings, and assault with a deadly weapon for threatening the woman in the car. By the time trial started, the prosecution had reduced the charges to first degree murder for the dead man, and three counts of assault with a deadly weapon. The prosecution tried to prove that he set out to kill the man who died, because Thompson thought the victim was a “snitch.”

Thompson testified. He did not deny the stabbings. His story was that it was a fight, and he stabbed the three men in self defense. He claimed that he thought the man he killed had a knife in his hand, because his own fist was cut when he hit the dead man’s fist. The other men, he claimed, were reaching behind their waists as if to get weapons. No weapons were found on the dead man or the two who were stabbed. A knife was found in *1573the woman’s car, in which the four victims had travelled to the fight.

The jury rejected the prosecution theory of the case, and convicted Thompson only of second degree murder for the killing, not first degree. It convicted him of assault with a deadly weapon for each of the nonfatal stabbings, but acquitted him of assault with a deadly weapon on the woman in the car.

The facts relevant to the issues on appeal relate to voir dire and closing argument. During voir dire, a venireman said he had read that Thompson had previously pleaded guilty to something and then withdrawn his plea. During trial, a witness “turned around” on the prosecution. The witness had apparently told the police that Thompson had set out to “get” the dead man because he was a “snitch.” The witness testified at trial that Thompson was just “backing up” the witness and got involved in the fight without intending to. In closing argument, the defense suggested that the’ police and prosecutor had pressured the witness to lie in his “snitch” statement. The prosecutor responded in her rebuttal argument by implying that she personally believed that Thompson was guilty of first degree murder, based on her own experience and the capabilities of the police. She suggested that the defense investigator had coached the witness to recant his “snitch” story, and had rehearsed him to testify favorably to Thompson.

Thompson appealed and exhausted his remedies in the state courts. He then petitioned for a writ of habeas corpus in federal district court. The district court denied his petition, and Thompson now appeals that denial.

II. Analysis.

We assume without deciding that the venireman’s disclosure fell within the category of juror misconduct, but decide that in the circumstances of this ease, the error was harmless. We also hold that the prosecutorial misconduct was not prejudicial.

A. Standard of Review.

We review de novo the district court’s denial of habeas relief. Dickson v. Sullivan, 849 F.2d 403, 405 (9th Cir.1988). For habeas review, Congress has directed that state court determinations of historical fact “shall be presumed to be correct.” 28 U.S.C. § 2254(d). However, “[t]he application of a legal standard to historical facts does not constitute a factual finding entitled to a presumption of correctness under section 2254(d).” Marino v. Vasquez, 812 F.2d 499, 504 (9th Cir.1987). Therefore we review the state court’s determination of lack of prejudice de novo. See Dickson, 849 F.2d at 405; Marino, 812 F.2d at 504.

B. Juror Misconduct.

Thompson had pleaded guilty to second degree murder, but had been allowed to withdraw his plea and go to trial. The voir dire took quite a long time. During the first two days of voir dire, the court had given admonitions not to read stories about the ease in the newspaper. The third day, a new venireman was brought in, who had not heard the admonitions. During the portion of voir dire by the court, the judge elicited that the venireman had read a newspaper story about the case the previous night. Defense counsel took the matter up during his own questioning, in open court with the other jurors present. The juror refrained from saying what the newspaper story had said the first two times he was asked. The third time he was asked whether he had read something about the case, the second time defense counsel had asked him, the venireman expanded on his previous two answers and said what he had read, that Thompson had pleaded guilty to something:

[Judge]: Do you know anything about this case?
A. Read about it in the paper, last night’s paper.
Mr. Perisho [defendant’s lawyer]: Mr. Fowler, you mentioned that you only recall this case from what you read in the papers. Did you say last night?
A. Yes.
Q. Did you read something about this case?
A. Yes, the jury selection pending on this case, pleaded guilty at one time and changed it.

*1574Thompson argues that this disclosure by the venireman amounted to prejudicial pretrial publicity. It does not. Prejudicial pretrial publicity may make it impossible to seat an impartial jury in a community, if the community has been saturated with prejudicial and inflammatory media publicity about the crime. See Jeffries v. Blodgett, 5 F.3d 1180, 1189 (9th Cir.1993); Harris v. Pulley, 885 F.2d 1354 (9th Cir.1988). A comment by a venireman during jury selection does not amount to pretrial publicity in this sense.

Thompson argued before the magistrate, and our dissenting colleague urges, that the venireman’s remark amounted to juror misconduct. Juror misconduct typically occurs when a member of the jury has introduced into its deliberations matter which was not in evidence or in the instructions. Marino v. Vasquez, 812 F.2d 499, 504 (9th Cir.1987). Thompson does not claim that any juror did anything wrong. We have used the sobriquet “jury misconduct” where no juror introduced the extraneous material, but a clerk inadvertently sent the police report and search warrant affidavit into the jury room. Hughes v. Borg, 898 F.2d 695 (9th Cir.1990).

The constitutionally protected interest at stake is in having the jury decide the case based on evidence subject to confrontation, cross examination, and the assistance of counsel. Dickson v. Sullivan, 849 F.2d 403 (9th Cir.1988). This interest requires inquiry for prejudice even where the extrinsic evidence comes from a source other than a juror, as when a deputy escorting the panel tells them that the defendant has “done something like this before.” Id.

We have found no cases of reversal because of “jury misconduct” or extraneous information, where the defense itself elicited the extraneous information in open court. If reversal on this ground were permitted, defense counsel could plant error in the record of any trial. The reasons for vacating convictions where the jury obtained extraneous information are protection of the right to confrontation, cross examination and counsel. These reasons do not apply where the information is elicited during voir dire, because defendant had counsel who cross examined the venireman in defendant’s presence. We assume, for purposes of discussion of our dissenting colleague’s views, but do not decide, that the venireman’s remark, “pleaded guilty at one time and changed it,” amounted to introduction of extraneous material on which reversal could be founded. We further assume without deciding that failure to grant a mistrial after the venireman’s remark was constitutional error.

The assumed error would support a writ only if prejudicial, because the error is “trial error,” not “structural error.” “Trial error ‘occur[s] during the presentation of the case to the jury’ and is amenable to harmless-error analysis because it ‘may ... be quantitatively assessed in the context of other evidence presented in order to determine [the effect it had on the trial].’ ” Brecht v. Abrahamson, 507 U.S. 619, 629, 113 S.Ct. 1710, 1717, 123 L.Ed.2d 353 (1993) (quoting Arizona v. Fulminante, 499 U.S. 279, 307-08, 111 S.Ct. 1246, 1264, 113 L.Ed.2d 302 (1991)). Deprivation of the right to counsel or a biased judge are “structural defects” because they infect the entire trial process. Fulminante, 499 U.S. at 306-10, 111 S.Ct. at 1262-65. “[M]ost constitutional errors can be harmless.” Id. at 306, 111 S.Ct. at 1263. The Supreme Court classifies admission of an involuntary confession as “trial error.” Id. at 306-11, 111 S.Ct. at 1262-66. In practical terms, admission of evidence of a withdrawn plea resembles admission of an involuntary confession, not a biased judge or total deprivation of the right to counsel. The purported error in the case at bar is in the exercise of discretion by a trial judge about how to cure prejudice to the defendant from information defense counsel elicited from a potential alternate juror. If the trial judge adequately cured the defect so that no actual prejudice resulted, the error is harmless and the defendant is not entitled to the writ.

The test for prejudice has recently been clarified by the Supreme Court. The standard for determining whether to grant the writ in collateral review of a state court conviction is not whether the error was harmless beyond a reasonable doubt, as in *1575Chapman v. California, 386 U.S. 18, 24, 87 S.Ct. 824, 828, 17 L.Ed.2d 706 (1967), but rather, whether the constitutional error “ ‘had substantial and injurious effect or influence in determining the jury’s verdict.’ ” Brecht, 507 U.S. at 623, 113 S.Ct. at 1713, quoting Kotteakos v. United States, 328 U.S. 750, 776, 66 S.Ct. 1239, 1253, 90 L.Ed. 1557 (1946); Jeffries v. Blodgett, 5 F.3d 1180, 1190 (9th Cir.1993).1

Burden of proof was clarified last term. The Court has “deliberately phrase[d] the issue ... in terms of a judge’s grave doubt, instead of in terms of ‘burden of proof.’” O’Neal v. McAninch, — U.S. -, -, 115 S.Ct. 992, 994-95, 130 L.Ed.2d 947 (1995). The reviewing judge should “ask directly, ‘Do I, the judge, think that the error substantially influenced the jury’s decision?’ ” Id. at -, 115 S.Ct. at 995. Only if the judge thinks the error substantially affected the jury’s decision, or the record is so evenly balanced that the judge is in “grave doubt,” should habeas relief should be granted. Hegler v. Borg, 50 F.3d 1472, 1478 (9th Cir.1995); United States v. Moorehead, 57 F.3d 875, 879 (9th Cir.1995).

The nature of the test for prejudice precludes categorical reversal, or affirmance, for introduction of extraneous information. In this case, we have reviewed the record de novo, and we do not think the venireman’s remark substantially affected the jury’s decision. Nor are we in the state of equipoise which would invoke the “grave doubt” rule.

The trial judge consulted extensively with counsel about how to cure the problem defense counsel had created. He decided on an admonition, carefully worded to prevent jurors from drawing inferences from what the juror had said, without reminding them of exactly what he had said:

Now ladies and gentlemen of the jury, just before we took the recess, while questioning was being done of Mr. Fowler, the proposed alternate juror, he was asked whether he had heard about this case or been told about this case and he mentioned that he had seen something in the Stockton Record. Again I want to caution all of *1576you that what that Stockton Record or any newspaper says or any other news media or television says is not in evidence in this case at all. You are not to consider it; you are not to discuss it; you are not to even think about it. You are not to read the newspapers concerning this incident or any other similar incident as long as this trial is in existence.

This admonition adequately told those jurors who understood from the veniremen that a newspaper reported Thompson had pleaded to something to disregard it and decide the case on the evidence. It also avoided educating those who had not heard or understood the alternate’s remark about the newspaper report.

The California Court of Appeals, the highest state court to review this case on the merits, decided that any inference of prejudice was rebutted by five things: (1) the juror’s own phrasing; (2) the judge’s admonition; (3) defense counsel’s failure to object to the phrasing of the admonition; (4) defense counsel’s failure to request individual voir dire of the jurors; and (5) defendant’s trial strategy of admitting the offenses but claiming self defense.

Here, a prospective alternate juror, later excused, stated he had read about defendant’s guilty plea in the newspaper. Assuming, arguendo, such a remark falls within the ambit of juror misconduct, raising a presumption of prejudice, .we hold such presumption rebutted by the record. The remark was vague and awkwardly phrased, and the jury, with defense counsel’s apparent concurrence, was generally admonished not to consider it. Defense counsel did not make any request for voir dire of the jurors. Moreover, as we have said, defendant’s strategy at trial was to admit the offenses but claim self-defense. On this record, we hold no prejudice actually resulted. (People v. Boyd, supra, 95 Cal.App.3d [577] at pp. 586-589 [157 Cal.Rptr. 293 (1979) ].)

People v. Thompson, 3 Criminal 12056, at 13 (Cal.Ct.App. Sept. 26, 1983). We agree.

The venireman’s remark that he read in the newspaper Thompson had “pleaded guilty at one time and changed it,” does not say to what charge Thompson had pleaded guilty. The jury was told not to use this remark against Thompson. Although in some cases it would have been hard for the jury to decide the case on any basis other than its knowledge about the withdrawn plea, in this case, there was no dispute about whether Thompson had stabbed the victims, just whether he had justification or mitigation. A withdrawn plea to an unknown charge did not give the jury answers to the questions upon which the case would probably be resolved. We can be confident that the jury did not ignore the evidence and just convict Thompson because it had learned of his prior plea, because the jury rejected the prosecution case for first degree murder, asked the judge serious questions about self defense, and deliberated for two full days and parts of two more before reaching its second degree murder verdict. The case is analogous to where a witness says something that was not supposed to be before the jury, the judge admonishes the jury to disregard it, and there is no reason to doubt that the jury obeyed the admonition.

C. Prosecutorial Misconduct.

We assume without deciding that the prosecutor’s remarks in her rebuttal argument amounted to misconduct. Even if a prosecutor’s argument is egregiously improper, a federal court cannot issue a writ of habeas corpus to state authorities unless “the prosecutors’ comments ‘so infected the trial with unfairness as to make the resulting conviction a denial of due process.’ ” Darden v. Wainwright, 477 U.S. 168, 181, 106 S.Ct. 2464, 2471, 91 L.Ed.2d 144 (1986) (quoting Donnelly v. DeChristoforo, 416 U.S. 637, 94 S.Ct. 1868, 40 L.Ed.2d 431 (1974)). The standard of review is the “‘narrow one of due process, and not the broad exercise of supervisory power.’ ” Id. “Improper argument does not, per se, violate a defendant’s constitutional rights.” Jeffries v. Blodgett, 5 F.3d 1180, 1191 (9th Cir.1993); see Duckett v. Godinez, 67 F.3d 734, 743 (9th Cir.1995).

We are not reaching the question, for which Brecht, 507 U.S. 619, 113 S.Ct. 1710, and O’Neal, — U.S. -, 115 S.Ct. 992, would supply the standard of review, of *1577whether the constitutional error if any was harmless. See Darden, 477 U.S. at 183 n. 15, 106 S.Ct. at 2472 n. 15 (“We do not decide the claim of prosecutorial misconduct on the ground that it was harmless error.... it was not constitutional error.”) We assume without deciding that the prosecutor’s argument was improper, and decide that the impropriety if any was not constitutional error. Only if the argument were constitutional error would we have to decide whether the constitutional error was harmless. A federal court does not have supervisory power, as a state appellate court might, to correct error not of constitutional dimension, and cannot issue a writ of habeas corpus where the improper conduct does not deprive the defendant of his constitutional right to due process of law. See Darden, 477 U.S. at 168, 106 S.Ct. at 2465.

The prosecutor’s argument in the case at bar did not so “ ‘infect[ ] the trial with unfairness as to make the resulting conviction a denial of due process.’ ” Darden, 477 U.S. at 181, 106 S.Ct. at 2471. We can be quite sure the jury’s immune system withstood the “infection,” because the prosecutor’s argument failed to persuade. The jury rejected her first degree murder argument. The jury’s rejection of the prosecutor’s argument shows that it knew better than to decide the case based on the prosecutor’s confidence in her first degree murder theory, and decided it instead based on its own view of the evidence.

AFFIRMED.

Notes

. Our dissenting colleague takes the position that we have ignored Dickson v. Sullivan, 849 F.2d 403 (9th Cir.1988), Jeffries v. Blodgett, 5 F.3d 1180 (9th Cir.1993), and Lawson v. Borg, 60 F.3d 608 (9th Cir.1995). Of course we have not. Our colleague would push error which he concedes was not ‘‘structural’' into a category of prejudice so strongly presumed to exist as to be indistinguishable as a practical matter from structural error.

Dickson, which came down before Brecht v. Abrahamson, 507 U.S. 619, 113 S.Ct. 1710, 123 L.Ed.2d 353 (1993), was not like this case. A deputy escorting the jury told the jurors out of court that the defendant had "done something like this before,” the lawyers and judge apparently did not discover this until after the trial was over, and the judge's boilerplate instructions, to decide the case on the evidence and disregard defendant’s prior convictions except as they might bear on credibility, did not cure the likely prejudice.

Jeffries, a death penalty case, also involved information conveyed outside the courtroom that the defendant was a convicted criminal, without the lawyers or judge knowing about the extraneous information and having an opportunity to try to cure unfair prejudice. We concluded that if the out-of-court communication to the jurors had been made, in the particular case it would have had a substantial and injurious effect. We particularly noted, in Jeffries, that under Brecht "petitioners now face a greater burden” than they had when we decided Dickson and Bayramoglu v. Estelle, 806 F.2d 880 (9th Cir.1986), and "must now show that the alleged error ‘had substantial and injurious effect or influence in determining the jury’s verdict.' " Jeffries, 5 F.3d at 1190.

In Lawson v. Borg, 60 F.3d 608 (9th Cir.1995), some jurors had told others that the defendant was known around town as a violent man. As in Dickson and Jeffries, the communication was outside the courtroom, unknown to the lawyers and judge, so, unlike the venireman's remark in the case at bar, the "exposure to facts not in evidence deprives a defendant of the rights to confrontation, cross-examination, and assistance of counsel.” Lawson, 60 F.3d at 612. We expressly applied the Brecht standard, noting that a habeas petitioner “is entitled to habeas relief only if it can be established that the constitutional error had 'substantial and injurious effect or influence in determining the jury's verdict.' " Id.

We have performed precisely the analysis which Brecht and our applications of Brecht in Jeffries and Lawson require. Our dissenting colleague would pigeonhole the error in this case into a general category, say that errors in that category are generally too prejudicial to be cured, and conclude that the error in the case at bar was prejudicial. That kind of pigeonholing and syllogistic inference from generalities is contrary to the particularized case-by-case analysis of prejudice required by Brecht.



Dissenting Opinion

REINHARDT, Circuit Judge,

dissenting:

The majority rides roughshod over long-established circuit precedent and abruptly discards our well-considered approach to cases in which the jury improperly receives highly prejudicial information. Worse, it treats the law of this circuit with such disregard that it pauses only briefly to discuss the controlling precedents, makes only the faintest attempt to distinguish them, and utterly fails to identify any legally significant basis for doing so. Because I find that the communication to the jury of information regarding Thompson’s withdrawn guilty plea deprived him of a fair trial, I would reverse the judgment of the district court, and would, therefore, not reach Thompson’s separate claim of prosecutorial misconduct.

I.

The majority can affirm the judgment of the district court only by disregarding the clear holdings of our prior cases that speak directly to the issue of juror misconduct or jury contamination, Dickson v. Sullivan, 849 F.2d 403 (9th Cir.1988), Jeffries v. Blodgett, 5 F.3d 1180 (9th Cir.1993), and Lawson v. Borg, 60 F.3d 608 (9th Cir.1995).1 The clear *1578import of those cases is that the exposure of the jury to highly prejudicial extraneous information 2 creates a rebuttable presumption of prejudice that, if unrebutted, is sufficient to satisfy the harmless error standard. The majority fails to address the highly prejudicial nature of information regarding a withdrawn guilty plea, as well as the strong presumption that such information has a substantial effect on the verdict. Moreover, it relies on factors that are patently insufficient under our precedent to rebut the presumption of prejudice. Under the controlling case law, there can be no doubt that the information communicated to the jury in this case was highly prejudicial and that the presumption of prejudice has not been rebutted. Accordingly, I must dissent.

Generally, a presumption of prejudice arises from the introduction of adverse extraneous evidence or information into the trial process.3 Dickson, 849 F.2d at 406. This is a rebuttable presumption, and we look to a variety of factors to determine whether the presumption has been rebutted in a particular case. See, e.g., Bayramoglu, 806 F.2d at 887 (listing five factors). Some of these factors are more important than others and not all are pertinent in every case.4

*1579Regardless of the particular factors examined, the ultimate question remains: What is the potential effect of the information upon the jury? Dickson, 849 F.2d at 406. That ultimate question is linked to the harmless error standard. When conducting a harmless error analysis in a habeas ease, we must ordinarily determine whether the error had a “substantial and injurious effect or influence in determining the jury’s verdict.”5 Jeffries, 5 F.3d at 1190 (quoting Brecht v. Abrahamson, 507 U.S. 619, 118 S.Ct. 1710, 123 L.Ed.2d 353 (1993)). Because whether there is a constitutional violation here depends on the likelihood that the extraneous information had a substantial and injurious effect on the verdict, the harmless error question is subsumed within the finding of a constitutional violation.6

The majority assumes that the information communicated to the jury here could warrant reversal but erroneously fails to consider the nature of the information. When assessing claims of prejudice in extraneous information cases, we place great weight on the nature of the information. Lawson, 60 F.3d at 612. We have consistently held that information regarding a defendant’s prior convictions is highly prejudicial. See, e.g., Dickson, 849 F.2d at 406; Jeffries, 5 F.3d at 1190. In Dickson, we explained, “Evidence of prior offenses is likely to prejudice a defendant even where the defendant has the opportunity to rebut the evidence and take other ameliorative steps.” Id. at 407. Moreover, where the prior offense is similar to the charged conduct in the case on trial, “the danger of prejudice to the defendant is exacerbated.” Id.

The prejudicial nature of a withdrawn guilty plea exceeds that of a prior conviction. The principal rationale behind classifying information regarding prior criminal acts as highly prejudicial is that there is an inevitable human tendency to conclude that a defendant who has once engaged in similar conduct is likely to have committed the specific similar act with which he is charged. See, e.g., Dickson, 849 F.2d at 407-08 (collecting and analyzing cases). An equally impermissible and even more injurious inference is likely to arise where the information concerns a plea of guilty, albeit subsequently withdrawn, in the very case before the jury. In such an instance, the jury learns that the defendant has admitted to the actual conduct currently being adjudicated. There is an “inevitable human tendency” to presume that innocent people do not plead guilty. Accordingly, the average person is likely to believe that an individual who pled guilty to the charged offense is indeed guilty, even if the defendant subsequently withdraws the plea. Such an inference, although natural, or “inevitable,” is intolerable because it is completely inconsistent with the “presumption of innocence” that is the starting point of every criminal trial. Information regarding the admission of the actual conduct at issue is thus inherently more prejudicial than information regarding conviction of a prior similar offense.

Moreover, we have expressly recognized that the publication of a defendant’s withdrawn guilty plea and excluded confession *1580can be “the most conclusively prejudicial publicity possible.” United States v. Polizzi, 500 F.2d 856, 881 n. 42 (9th Cir.1974). Indeed, the trial court recognized this during an in limine hearing prior to the selection of the jury when the prosecutor indicated her desire to admit into evidence Thompson’s withdrawn guilty plea. The court, refusing the prosecutor’s request, stated, “You may like to use it. I can tell you right now I will not permit it.... That would be highly prejudicial- That would be terribly prejudicial.” RT 453:28; 454:1-15. If this information was “highly prejudicial” before jury selection, it certainly did not become less so after the jury was sworn. See Gibson v. Clanon, 633 F.2d 851, 855 (9th Cir.1980) (fact that extraneous information was previously ruled inadmissible is a relevant factor in determining prejudicial effect on jury); see also Hughes, 898 F.2d at 700, 701 (extraneous information is less likely to be prejudicial where material is merely duplicative of evidence properly introduced in open court).

Thus, had the majority engaged in the requisite analysis of the nature of the information communicated here to the jury, it would have found not only that the information is highly prejudicial, but that it is considerably more prejudicial than that communicated in Dickson or Jeffries. Not only is information regarding a withdrawn guilty plea more inflammatory than information regarding a prior conviction, but a withdrawn plea necessarily has a more direct bearing on material issues in the case. The ultimate question to be decided by the jury is the defendant’s guilt or innocence. A plea of guilty to the charges being adjudicated provides a complete answer to that ultimate question.

Having ignored the highly prejudicial nature of the information, the majority then disregards the strong presumption of prejudice that exists here. Indeed, where information concerning prior convictions is concerned, we have found it most unlikely that the presumption of prejudice could be rebutted. See Dickson, 849 F.2d at 407 (collecting cases). In Jeffries, for example, we concluded that the extraneous information communicated was so highly inflammatory and prejudicial that such information would necessarily have a “substantial and injurious effect or influence on the verdict” sufficient to satisfy the harmless error standard. 5 F.3d at 1191. Because the information communicated to the jury in this case was more prejudicial than that in Dickson and Jeffries, the presumption that the jury’s verdict was affected is even stronger, and it is accordingly even less likely that the presumption could be rebutted.

Last, without considering any of our precedent, the majority cursorily concludes that any presumption of prejudice was rebutted. Here, however, as in Jeffries and Dickson, the factors we have previously identified are insufficient to rebut the presumption.7 Moreover, under circuit precedent, the existence of the four circumstances identified by the California Court of Appeals and accepted by the majority is wholly insufficient to rebut such a presumption in this ease. See, e.g. Jeffries, 5 F.3d at 1190; Dickson, 849 F.2d at 406-07. Those four circumstances are: 1) the way in which the remark was phrased; 2) the fact that a general admonition was given; 3) the absence of a request to voir dire the jurors; and 4) the nature of the defense.

I have already discussed the first circumstance — the nature of the remark. It is highly prejudicial. I underscore, however, that the state court of appeals found as a fact that the prejudicial answer was given in the presence of the jury and that the answer communicated the fact that defendant had pleaded guilty, albeit vaguely.

Nonetheless, the state court found, and the majority agrees, that any prejudice flowing from the fact that the jurors learned that the defendant had pled guilty was cured by the second circumstance — the trial court’s subsequent general admonition to the jury. As a general rule, “A timely instruction from the judge usually cures the prejudicial impact of evidence unless it is highly prejudicial or the instruction is clearly inadequate.” Bayramoglu v. Estelle, 806 F.2d 880, 888 (9th Cir.1986) (emphasis added) (quoting United States v. Berry, 627 F.2d 193, 198 (9th Cir. *15811980)). Here, the admonition falls within both exceptions to the curative rule: the information was extremely prejudicial and the admonition inadequate.

This case is quite similar to Dickson. There, we rejected a similar argument regarding a general admonition, stating:

Although we ordinarily assume that instructing the jury to disregard extraneous evidence sufficiently ensures that inadmissible evidence will not influence the jury, where the extrajudicial statement concerns a defendant’s prior criminal acts, the efficacy of such instructions is subject to serious doubt.

849 F.2d at 408 (internal citations omitted) (emphasis added). Then-Judge (now Chief Judge) Wallace, writing for the court, explained that instructing a jury to ignore prior acts information in judging the offense at trial “is to ask human beings to act with a measure of dispassion and exactitude well beyond mortal capacities.” Id. at 408 (quoting United States v. Lewis, 787 F.2d 1818, 1323 (9th Cir.1986) (quoting United States v. Daniels, 770 F.2d 1111, 1118 (D.C.Cir.1985))). Even more so with a guilty plea, withdrawn or not. Given the highly prejudicial nature of the information communicated, the court’s admonition failed to cure the prejudicial effect.

As to the third circumstance — an absence of a request for voir dire — I note that a trial court has an affirmative obligation to investigate potential bias within the jury. See, e.g., Smith v. Phillips, 455 U.S. 209, 212-220, 102 S.Ct. 940, 944-47, 71 L.Ed.2d 78 (1982) (Due Process means “a trial judge ever watchful to prevent prejudicial occurrences and to determine the effect of such occurrences when they happen.”); see also Reiger v. Christensen, 789 F.2d 1425, 1434 (9th Cir.1986). Thus, if a special voir dire was required in order to attempt to eliminate any prejudice, then the trial judge had a sua sponte duty to conduct it. Here the trial judge was leery of taking steps which would highlight the comment in the minds of the jurors and possibly exacerbate its prejudicial effect. I need not reexamine that conclusion now, but I do note that refraining from taking a step that may increase the likelihood of prejudice does nothing to rebut a presumption of prejudice.

Finally, as to the fourth circumstance, I simply do not see how the fact that defendant asserted a self-defense theory at trial removed the taint of prejudicial information of a withdrawn guilty plea. A claim of self-defense is nonetheless an attempt to deny legal culpability or “guilt.” An admission of guilt is as inconsistent with that defense as with any other.

Thus, I would conclude that, in light of all the circumstances, the nature of the information communicated here was so highly prejudicial as to compel the conclusion that a constitutional violation occurred. The serious concerns that led us to grant habeas relief in Dickson and Jeffries are present here and lead ineluctably to the conclusion that the extraneous information communicated to the jury in this case had a substantial and injurious effect on the verdict. See Jeffries, 5 F.3d at 1191. Only by discarding our settled approach to the introduction of extraneous information can the majority reach a contrary conclusion.

The majority responds in footnote 1 by attempting to distinguish this case from Dickson, Jeffries, and Lawson, on the ground that the communications in those cases, unlike the one here, were made outside the courtroom and were unknown to the lawyers and judge. While it offers this purported distinction with much bluster, it has no force whatsoever. However true it may be that the communication here was in the courtroom and was known to the lawyers and judge, that fact is wholly irrelevant. As the majority says that it recognizes, it is “[j]ury exposure to facts not in evidence” that deprives a defendant of his Sixth Amendment rights. Lawson, 60 F.3d at 612. Thompson’s withdrawn guilty plea was exposed to the jury and not introduced into evidence. That squarely meets the requirements the majority acknowledges are applicable. Moreover, as I noted earlier, Thompson’s withdrawn guilty plea was excluded precisely because the court concluded that its introduction would be “terribly” prejudicial. Nevertheless, that terribly prejudicial information was disclosed to the jury in the courtroom. *1582Ah, the majority says, but in our other eases the disclosure did not occur in the courtroom. Neither, I might observe, did it occur during the month of April.

Would it have been more harmful if the highly prejudicial information had been disclosed to the jurors in the water closet? the hallway? the subway? I would think not — in fact it might have been less harmful. In this case, the jurors were all assembled and they were all told of the highly damaging information at the same time. In the other cases, only some of the jurors ever heard the harmful information.8 That, in this case, the lawyers and the judge also heard it makes no difference, unless, as a result, they took some effective action that served to cure the error. As I have explained earlier, no such adequate curative action was taken. Here, as in our prior cases, the jury was exposed to highly prejudicial “facts not in evidence” — facts that were more harmful than the facts communicated in any of our earlier cases. That is all that we need to know in order to reach our decision.

None of the cases cited by the majority placed any significance on where the information was communicated to the jurors. However, if one wished to create a distinction, it would be the converse of the one the majority purports to draw here. Information conveyed in the courtroom is generally understood to be more authoritative and thus more damaging than information that is conveyed in any other setting — not less. Faced with a distinction as whimsical, indeed as perverse, as the one the majority makes here, there can be no question that my colleagues have simply disregarded the controlling case law.

The majority’s claim that I have “pushed” the error in this case into a category in which prejudice is “strongly presumed,” contrary to the case-by-case analysis required by Brecht, serves only to demonstrate, once again, its disregard of binding Ninth Circuit precedent, precedent that applied Brecht in the precise manner I do. In Jeffries and Lawson, we applied the rebuttable presumption of prejudice established in Dickson, placing “great weight on the nature of the extrinsic evidence introduced.” Lawson, 60 F.3d at 612; see Jeffries, 5 F.3d at 1190. In each case, applying the Brecht standard, we concluded that because the extraneous information “was both directly related to a material issue in the ease and highly inflammatory,” and the presumption of prejudice was not overcome, the information “substantially and injuriously influenced the verdict.” Lawson, 60 F.3d at 613; Jeffries, 5 F.3d at 1190-91. In this case, where the extraneous information was directly related to a material issue— in fact, it was related to the material issue— and was highly inflammatory — even more inflammatory than the information communicated in Dickson, Jeffries, and Lawson — we are obligated to reach the same conclusion. Thus, if I have “done” anything to the error in this ease, it has been to assess it under the principles established by controlling Ninth Circuit precedent, rather than go along with a jury-rigged, highly-puzzling rule that conflicts squarely with the law of this circuit.

II.

Although I would conclude that the communication of highly prejudicial extraneous information to the jury infringed Thompson’s constitutional right to a fair trial and end the analysis here, the state’s final argument remains: that the error should be deemed harmless because the evidence against Thompson was overwhelming. Our precedent, contrary to the suggestion by the majority and consistent with relevant Supreme Court authority, see n. 5, supra, do not require a further search for “actual prejudice” after we find a constitutional violation in extraneous information eases.9 For when we *1583find such a violation, we have already concluded that the Brecht test has been met. Having performed a prejudice analysis and having found the extraneous information highly prejudicial before determining that a constitutional violation occurred, it would be both anomalous and duplicative to perform a second prejudice analysis.

Nevertheless, in a recent case, when faced with an argument by the state that the error was harmless due to the overwhelming evidence of guilt, we elected to answer the state’s argument on the merits. See Lawson, 60 F.3d at 613. Although Lawson relied on Jeffries and Dickson in arriving at the conclusion that prejudicial error had occurred, when analyzing the state’s harmless error argument, we did not discuss the relationship between our extraneous information doctrine and the harmless error inquiry. Perhaps because we found the state’s argument so clearly meritless, we concluded that such a discussion was unnecessary. In any event, without suggesting that we were required to address the state’s “alternative” argument, we simply rejected it summarily on the facts. Id. at 613.

Although not required to do so under our precedent, I think it informative to follow a similar course in this case. A review of the record reveals that, at the very least, there is grave doubt as to whether the information substantially affected the verdict. I have no doubt that Thompson’s ease was a close one. The majority’s conclusion that the information communicated did not substantially affect the verdict is belied by the facts. Because the majority only cursorily describes those facts, I will briefly provide the necessary context for understanding the highly prejudicial nature of the information communicated to the jury.

The conflicting theories and evidence at trial presented a difficult case for the fact finder.10 Thompson never denied stabbing the Valverdes. The pivotal jury issue in the case was of what degree of homicide, if any, was the defendant legally culpable. The conflicting testimony and evidence presented at trial did not clearly point to an answer. The issue was largely one of credibility.

Moreover, the case was complicated by the dispute during the trial and closing arguments regarding a prior conversation between the defendant and the prosecutor. This dispute revolved around when the defendant’s self-defense theory developed and whether the prosecutor originally assured the defendant that the events only suggested a manslaughter charge. This disputed eon-*1584versation was heavily argued before the jury. As a result, the credibility of the defendant and the prosecutor were directly pitted against each other, and this dispute became the center of a significant portion of the prosecution’s closing argument.11

The jury deliberated for three days and continuously grappled with the self-defense issues — issues central to Thompson’s primary defense theory — sending numerous questions and requests for assistance to the trial court.12 During the second day of deliberations, the jury found Thompson guilty of one of the assault with a deadly weapon charges but acquitted him of a second such charge. Late on the third day of deliberations, after seeking additional clarification and instructions from the court,13 the jury finally returned a verdict finding Thompson guilty of second degree murder and of an additional assault with a deadly weapon charge. The verdict of second-degree murder represented a rejection of the prosecution’s ultimate theory of the case because it negated a finding of the premeditation required for the prosecution’s request for first-degree murder. However, it also represented a rejection of Thompson’s self-defense arguments which would have entitled him either to an acquittal or a manslaughter verdict for “imperfect” self-defense.

Based on this record, I cannot share the majority’s confidence that the verdict was not substantially affected by the information communicated to the jury. Given the conflicting testimony and evidence, and given the importance of credibility to the jury’s determination, I cannot conclude that the introduction of information regarding Thompson’s withdrawn guilty plea was harmless. Consistent with the principles developed in Dickson, Jeffries, and Lawson, I am compelled to conclude that the decision of the district court should be reversed. Accordingly, I would remand with instructions to issue the writ of habeas corpus unless the state of California elects to grant a new trial within 30 days of the issuance of the mandate.

. I believe that the majority need not equivocate as to the framework to be applied in this case. Although the petitioner originally characterized his claim as one of "juror misconduct,” the district court analyzed it as a "pretrial publicity” case. I think that the district court erred and that the analysis originally proffered by the petitioner reflects the correct approach.

The pretrial publicity framework is clearly a poor fit for this case. Prejudicial publicity arguments are raised ordinarily in the context of a request for a change of venue. Typically, the defendant asserts that the media coverage relating to the crime prejudices his ability to obtain an impartial jury. See, e.g., Jeffries, 5 F.3d at 1188-89; Harris v. Pulley, 885 F.2d 1354, 1361-64 (9th Cir.1988). When analyzing a prejudicial publicity claim on habeas review, we examine the volume, content, and timing of news reports to determine the effect of the pretrial publicity upon the defendant’s ability to obtain an impartial jury. Jeffries, 5 F.3d at 1189. A key factor in this analysis is whether a high percentage of veniremen admitted to a previously formed opinion on the case. Harris, 885 F.2d at 1364. We also examine the quality and extent of voir dire and the responses obtained from seated jurors to determine whether their impartiality was established. Id.

Here, the parties are essentially arguing about the prejudicial impact of one statement that was made before a group of jurors who had already been sworn. Accordingly, we should concentrate on the nature and effect of that statement rather than on the extent of media coverage, the quality of voir dire for prospective jurors, and the percentage of the community infected by prejudicial publicity. We normally follow this approach when we review errors involving the communication of extraneous information or evidence to the jury. We do so under the rubric of "jury misconduct,” although the nomenclature is ill-suited for many of the cases that fall within the category. See, e.g., Jeffries, 5 F.3d at 1189-91 (one juror informed others that defendant was a convicted armed robber); Hughes v. Borg, 898 F.2d 695 (9th Cir.1990) (police report not admitted into evidence inadvertently left in jury room); *1578Dickson, 849 F.2d 403 (deputy sheriff acting as bailiff told two jurors about defendant's prior criminal conduct). Jury contamination would be a far more appropriate designation. In any event, I think it clear that what we have historically called the juror misconduct mode of analysis, rather than the pretrial publicity framework, is appropriate for use in this case.

Although I believe that the juror misconduct form of analysis must be followed here, this case differs in one significant respect from the other extraneous information cases we have considered. Once a trial is underway and testimony has been taken, there is an interest on the part of both parties and society generally in having the trial run its course. Accordingly, the prejudicial nature of the extraneous information must be fairly significant in order to warrant declaring a mistrial. In contrast, here, the eviden-tiary proceedings had not commenced; all that had occurred at the time the jury was contaminated was the initial stage of voir dire and the swearing in of the regular jurors. It would have been relatively easy and painless at that point to dismiss the jurors who had been selected and recommence voir dire with a new group of venireman. Thus, it can be argued that when considering the need for a mistrial at this stage of the proceedings, the prejudicial nature of the information should be measured by a lesser standard than the one we traditionally impose in extraneous information cases. However, I would not decide that question here because even under a traditional extraneous information analysis reversal is clearly required in this case.

.For the sake of clarity, I use the term "extraneous information” throughout, although judicial opinions use the term interchangeably with the terms "extrinsic information," "extrinsic evidence,” and "extraneous evidence.” I can perceive no difference in meaning in the way in which these terms are used in cases of the type before us. Compare Lawson, 60 F.3d at 612 (using both "extrinsic evidence” and "extrinsic information”) and Jeffries, 5 F.3d at 1190 (same) with Dickson, 849 F.2d at 407 ("extraneous evidence”). Any one of the terms would be satisfactory for purposes of this case.

. The majority argues that the extraneous information in this case was elicited by the defense, and states that to grant reversal where the defense itself has elicited the extraneous information would permit defense counsel to plant error in the record of any trial. Ante at 1574. The majority, however, cannot legitimately question counsel's intentions in this case. The record does not support the conclusion that defense counsel's questioning was designed to plant error or sabotage the trial. Nothing in the record suggests that counsel knew the contents of the story that the alternate juror had read. It is clear from an earlier admonition that the court issued to the seated jurors that the locality at issue was served by five separate publications. There is no reason to presume that defense counsel was aware of the precise contents of each issue of every one of these publications.

Moreover, although the trial court initially rebuked defense counsel for his poor phrasing of the voir dire question, he deliberately refrained from accusing counsel of attempting to cause a mistrial. Even more compelling is the California Court of Appeals’ treatment of this issue. That opinion is completely devoid of any reference to the "fault” or “intent” of defense counsel. Not one word of the opinion even arguably suggests that defense counsel deliberately or inadvertently created the cause for a mistrial. This complete silence regarding what would otherwise be a significant factual question — attempted sabotage — shows that the majority’s suggestion that error might have been planted intentionally is wholly inappropriate in this case.

. Among the factors we sometimes assess are; how the extraneous information was received; whether it was introduced prior to reaching a verdict; the length of time between introduction of information and deliberations; the significance of the nature of the extraneous information; whether it related to a material issue in the case; the inflammatory effect of the information; *1579and other factors that reasonably bear on the potential for prejudice. See, e.g., Jeffries, 5 F.3d at 1190-91.

. There are some errors that require a specific, or an actual, showing of prejudice to be reversible; however, there are some errors which are reversible because they are presumed to be prejudicial (and the inference has not been rebutted) and some which fall into a special category of errors that are reversible without regard to prejudice. See, e.g., United States v. Olano, 507 U.S. 725, 734, 737-41, 113 S.Ct. 1770, 1778, 1780-81, 123 L.Ed.2d 508 (1993) (reviewing a direct appeal under plain error standard); Lee v. Marshall, 42 F.3d 1296, 1298-99 (9th Cir.1994) (reviewing a habeas case under harmless error standard). The third kind, the "special category" cases, are commonly referred to as "structural defect” cases. This is not a structural defect case. Rather, under our circuit precedent, the error at issue is of the kind that is presumed to be prejudicial. The majority, however, ignores this fact and erroneously suggests that the error at issue requires a showing of actual prejudice.

. In this respect, the analysis is similar to that made in the case of an alleged Brady violation. There is no Brady violation unless the government fails to disclose evidence and "there is a reasonable probability that, had [it] been disclosed to the defense, the result of the proceeding would have been different." United States v. Woodley, 9 F.3d 774, 777 (9th Cir.1993). No second and separate actual prejudice review is undertaken.

. See, e.g., Jeffries, 5 F.3d at 1190-91; Bayramo-glu, 806 F.2d at 887. See also n. 4, supra.

. I recognize, of course, that "even a single juror's improperly influenced vote deprives the defendant of an unprejudiced, unanimous verdict." Lawson, 60 F.3d at 613. I note that all of the jurors were exposed to the highly prejudicial information only to make the point that if there is any distinction to be drawn between this case and our prior cases, it should militate in favor of reversal because not only was the information more prejudicial but more jurors were exposed to it, thereby even more clearly depriving Thompson of a fair trial.

. The rule established in Dickson and Jeffries is that where the extraneous information is “highly prejudicial” and the presumption has not been *1583rebutted, the harmless error standard is met without requiring a further showing of actual prejudice. Indeed, the absence of a further actual prejudice inquiry, was one of the criticisms lodged by the dissent in Jeffries. See Jeffries, 5 F.3d at 1198-99 (Fernandez, J., dissenting). The dissent summarized its dispute with the majority as follows:

The question with which we are presented is a narrow one: if a juror has heard [extraneous prejudicial information], is that juror so tainted that the verdict must be overturned? ... The majority, relying on Dickson and on its own independent reasoning, says that the answer is “yes” and when faced with those two facts sweeps away all other considerations. I say, not necessarily, and that we must still decide whether there was an effect on the verdict under all of the circumstances of this case.

Id.

. At trial, the state and Thompson presented wildly divergent theories of the brawl at Church’s and of Thompson’s degree of legal culpability. Thompson asserted that his friend Huerta had a long standing disagreement with the Valverde group. Thompson testified that when he and his friends encountered Tino’s group near the restaurant, he was asked to “back-up” Huerta in case there was trouble. When the fight erupted, only fisticuffs were involved at first. However, Thompson testified that during the fight he thought he saw some of the opposition reach for weapons, and he felt a slash on his hand. He immediately pulled out his knife and stabbed at his adversaries. (Although other witnesses told police that the Val-verde group was unarmed, a long kitchen knife was later found in their car, which had been present at the scene.) In sum, Thompson asserted that he had acted in self-defense.

In contrast, the prosecution pursued a theory of premeditated murder. It asserted that Thompson had sought Tino out to kill him in retaliation for Tino's adverse testimony at a trial of one of Thompson’s associates. Although, during a long intense police interrogation, Huerta made statements consistent with the prosecution’s theory (e.g. he said that Thompson had referred to Tino as a "snitch”), when called by the prosecution at trial, Huerta recanted his pri- or statements and corroborated Thompson's version of events.

. This dispute also led to the prosecutorial misconduct claim.

. On the second day of deliberations, the jury at various times requested more information on mutual combatants and self defense; requested copies of jury instructions on that issue; and specifically asked "Does the right of self-defense cease to exist when a person agrees to backup another person in a fight?”

.The jury sought more information relating to different degrees of homicide and specifically asked to see the district attorney's chart related to this issue. The jury again asked questions regarding the legal implications of agreeing to watch someone’s back, and whether and when someone would lose his right to self-defense.