Opinion for the Court filed by Chief Judge MIKVA.

MIKVA, Chief Judge:

Appellant submitted a Freedom of Information Act (FOIA) request for the results of telephone polls conducted by outside pollsters hired by the Air Force. See 5 U.S.C. § 552. Although the Air Force had released some of the poll results on its own initiative, the Air Force refused to turn over the vast majority of the information requested by appellant, claiming that release of the material would threaten its “deliberative process.” The district court agreed with the Air Force and held that the Air Force was entitled to withhold all of the surveys pursuant to Exemption 5 of FOIA. 5 U.S.C. § 552(b)(5).

Despite appellant’s specific request, the district court failed to enter a finding of segregability. This error warrants a remand. See PHE, Inc. v. Dept. of Justice, 983 F.2d 248, 252 (D.C.Cir.1993). The poll results released voluntarily by the Air Force contain purely factual information which could not threaten the Air Force’s deliberative process in any way. Yet the affidavits submitted by the Air Force in support of its refusal to disclose do not even hint that the poll results withheld are different from those released in any relevant respect. While the Air Force has not “waived” its right to claim an exemption from disclosure simply because it has released information similar to that requested, Abbotts v. NRC, 766 F.2d 604, 607 (D.C.Cir.1985), the fact that some of the information in the surveys is completely harmless suggests that other information in the surveys also might be released without threatening the Air Force’s deliberative process. The Air Force has the burden of demonstrating that no reasonably segregable information exists within the documents withheld. PHE, Inc., 983 F.2d at 252. That burden has not been met here. We therefore reverse the district court’s decision and remand for an express finding of segregability-

Appellant is the Army Times Publishing Company, publisher of several periodicals, including Air Force Times. Air Force Times is a weekly publication that contains news stories, features and opinions of inter *1069 est to present and former members of the Air Force. In 1983, Air Force Times submitted a FOIA request for the results of surveys conducted by the Air Force through the “Computer Assisted Telephone Interview Program” (“CATI”). Under this program, the Air Force retained a civilian market research firm to conduct a series of telephone polls of randomly selected groups of approximately 400 to 1,600 officers and enlisted personnel. Each survey respondent was asked a series of questions about working conditions within the Air Force, such as pay, bonuses, training and commissaries, and the results were tabulated.

The Air Force has released some CATI survey results voluntarily. For example, on August 7, 1990, the Air Force issued a news release reporting that ninety percent .of survey respondents believe that “Air Force Commissaries are heading in the right direction,” particularly with respect to “low prices,” “quality produce” and “cleanliness.” News Release #69-70, United States Air Force, August 7, 1990. The news release also reported the answers to other survey questions, including how much money respondents spent at the base commissary, how often and on what days they shopped there, and how satisfied they were with the commissary compared to civilian alternatives.

In its FOIA request, Air Force Times sought the disclosure of all the CATI survey results that had not been released voluntarily by the Air Force. At the urging of the Air Force, Air Force Times narrowed its FOIA request to eight specific topics such as pay, training and evaluation. The Air Force identified six thousand pages responsive to the request, but released only seven pages. The Air Force asserted that the remaining material was exempt from disclosure pursuant to Exemption 5 of FOIA, 5 U.S.C. § 552(b)(5). After exhausting its administrative appeal, the Army Times Publishing Company brought this action in district court to compel disclosure of the survey results. For the sake of clarity, we will refer to appellant by the name of the FOIA requester, Air Force Times.

The Air Force submitted a Vaughn index to the district court purporting to describe the withheld material and justify its refusal to disclose. See generally, Vaughn v. Rosen, 484 F.2d 820 (D.C.Cir.1973), cert. denied, 415 U.S. 977, 94 S.Ct. 1564, 39 L.Ed.2d .873 (1974). In its motion for summary judgment, the Air Force argued that the randomly selected survey respondents were participants in the deliberative process of agency policy-making, and therefore their “opinions and recommendations,” even when tabulated in aggregate form, were exempt from disclosure under Exemption 5. The Air Force based its claims on two affidavits by Judy K. Roomsburg, a “Behavioral Analyst” involved in the CATI programs.

The district court granted summary judgment for the Air Force. The court determined that the surveys were integral to the policymaking process within the Air Force, and that the survey respondents would be less likely to express their candid opinions if they were not assured that the aggregate poll results would be kept confidential. Although the Air Force Times specifically requested the court to determine whether there was at least some non-exempt material in the surveys, similar to that already released, that was segregable, the district court declined to make the requested segregability finding and sustained the Air Force’s blanket claim of exemption.

II.

Exemption 5 of FOIA states that an agency is not obligated to disclose “inter-agency or intra-agency memorandums or letters which would not be available by law to a party other than an agency in litigation with the agency.” 5 U.S.C. § 552(b)(5). Congress enacted this exemption largely to ensure that agencies “not lose the protection traditionally afforded through the evidentiary privileges simply because of the passage of FOIA.” Coastal States Gas Corp. v. Dept. of Energy, 617 F.2d 854, 862 (D.C.Cir.1980). The “deliberative process” privilege is central among the privileges protected by Exemption 5. However, as with all exemptions under FOIA, the deliberative process privilege must be construed as narrowly as is consistent with efficient government operation. Wolfe v. HHS, 839 F.2d 768, 773 *1070 (D.C.Cir.1988) (era banc). “Disclosure, not secrecy, is the dominant objective” of FOIA’s statutory scheme. Department of Air Force v. Rose, 425 U.S. 352, 361, 96 S.Ct. 1592, 1599, 48 L.Ed.2d 11 (1976).

In order for a document to be protected by the deliberative process privilege, it must be predecisional and deliberative, reflecting the advisory and consultative process by which decisions and policies are formulated. Wolfe, 839 F.2d at 774. Exemption 5 “rests most fundamentally on the belief that were agencies forced to ‘operate in a fishbowl,’ the frank exchange of ideas and opinions would cease and the quality of administrative decisions would necessarily suffer.” Dudman Communications Corp. v. Dept. of the Air Force, 815 F.2d 1565, 1567 (D.C.Cir.1987) (quoting S.Rep. No. 813, 89th Cong., 1st Sess. 9 (1965)).

The predecisional character of the material at issue here is not in dispute. Both sides acknowledge that the surveys were initiated to help the Air Force develop appropriate policies and determine how best to implement them. The propriety of the application of Exemption 5 thus turns on whether the withheld documents were correctly characterized by the Air Force and the district court as deliberative.

The Air Force Times emphasizes that it does not seek to learn the identity of any survey respondent or the responses of any particular individual. Rather, its request is only for the aggregate results of the surveys, including what number and percentage of respondents gave each answer and the methodological information necessary to interpret the results. The Air Force Times argues that this type of aggregate statistical information is “not so candid or personal in nature that public disclosure is likely in the future to stifle honest and frank communication within the agency.” Coastal States, 617 F.2d at 866.

In addition, the Air Force Times argues that the decision by the Air Force to release selected poll results from similar surveys is fundamentally inconsistent with its current claim that release of the poll results would threaten the decisionmaking process of the agency. In the Air Force Times’ view, the Air Force basically asserts a right to release those surveys that reflect favorably on the Air Force, and withhold those that are not so favorable, without regard to whether disclosure would actually interfere with the Air Force’s deliberative process.

The Air Force responds by predicting a total collapse of the CATI program if survey results are released under FOIA. This claim rests entirely on the affidavits of Major Roomsburg. Parroting the case law, Major Roomsburg states in her first affidavit that “[a]ny disclosure of the information withheld would impair the deliberative process of the Air Force by inhibiting full and frank exchange of views necessary with respect to such matters.” She later adds that “[t]he protection of CATI data encourages frank and open responses between subordinates and superiors at every level of command.” In her second affidavit, Major Roomsburg elaborates a bit, although just barely: “If the respondents are told (which they would have to be) that their answers will be released under FOIA.... the rate of refusal to participate would go up considerably [and] ... respondents would likely provide guarded responses or even alter their responses to reflect what they believe the Air Staff would want to hear.”

Thus, the Air Force’s theory boils down to the conclusory assertion that survey respondents would be more likely to temper their responses or refuse to participate if they knew that the aggregate results of the survey could be released to the public. This is an interesting claim, indeed, given that the Air Force, on its own initiative, selectively releases aggregate survey results to the public from time to time. Perhaps most remarkable is the fact that the Air Force routinely provides the survey results to the command of the Air Force, the very group that would have the most cause to resent critical answers and the greatest ability to retaliate. Under the CATI program, even the names of individual respondents are provided to Air Force supervisors. This aspect of the CATI program renders Major Roomsburg’s last allegation-that release of the poll results under FOIA would lead respondents to “alter *1071 their responses to reflect what they believe the Air Staff would want to hear” — makes little, if any, sense. Whatever chilling effect may be caused by release of the survey results to Air Force brass is already present. Moreover, the Air Force Times does not seek the answers of individual respondents. The Air Force Times seeks only the aggregate results of the surveys — some of which have already been released by the Air Force.

The Air Force retorts that release to Air Force supervisors is far different than release to the public. According to the Air Force, respondents are more likely to fear public reaction to the release of aggregate poll results than the reactions of Air Force supervisors to individual responses. The Air Force also claims that its “past discretionary releases of other information are simply not relevant,” because an agency does not waive its right to assert an exemption by releasing information that is only similar in nature to the requested material. See Abbotts v. NRC, 766 F.2d 604, 607-608 (D.C.Cir.1985). The Air Force explains that it merely determined that some survey results were “innocuous enough” that release “would not discourage respondents from participating in future surveys.” Appellee’s Brief at 17-18.

While it is true that the Air Force has not “waived” its right to claim an exemption from disclosure simply because it has released information similar to that requested, the fact that some of the information in the surveys is plainly factual and poses no threat to the agency’s deliberative process suggests that other information in the surveys could also be released. By releasing certain poll results and withholding others, the Air Force itself has demonstrated that all the surveys, taken together, are not worthy of a blanket claim of privilege under Exemption 5. This fact is critical, since Exemption 5 applies only to the deliberative portion of a document and not to any purely factual, non-exempt information the document contains. Coastal States, 617 F.2d at 867. Non-exempt information must be disclosed if it is “reasonably segregable” from exempt portions of the record, see 5 U.S.C. § 552(b), and the agency bears the burden of showing that no such segregable information exists. PHE, Inc., 983 F.2d at 252. Where non-exempt information exists within a document, “a district court clearly errs [by approving] a government’s withholding of information under FOIA without making an express finding of segregability.” Id. at 252 (citing Schiller v. NLRB, 964 F.2d 1205, 1210 (D.C.Cir.1992)).

The Air Force urges us to excuse the district court’s failure to enter a segregability finding on the grounds that its Vaughn index is sufficiently detailed to demonstrate that no reasonably segregable information has been withheld. Yet the Vaughn index submitted by the Air Force is hardly as specific as the Air Force claims. Nothing in the index or Major Roomsburg’s affidavits suggests that the withheld information is different in any relevant respect from that which has been released voluntarily. In fact, many of the entries in the Vaughn index contain no description of the listed document whatsoever. As we explained recently in PHE, Inc.: “Because only the agency knows the substance of the withheld information, the agency affidavits have immense significance in a FOIA case. Both the court and the requester must look to the affidavits for an explanation of an agency’s decision to withhold information.” Id. at 250.

The affidavits submitted in this case do not even hint that the withheld surveys touch upon matters any more delicate than the respondents’ use of their local commissaries. Of course, it is possible that the poll results do contain information of a more sensitive nature — information which, if released, would interfere with the Air Force’s deliberative process. Unfortunately, upon this record, it is impossible to discern a basis for distinguishing between the “innocuous” material and that which is potentially harmful. Without greater specificity from the Air Force in its description of the withheld documents, there is simply no means by which we may determine whether disclosure of the requested material would actually “discourage candid discussion within the agency.” Dudman, 815 F.2d at 1568.

In sum, the failure of the Air Force to offer some distinguishing feature of the withheld information strongly suggests that at least some of the information contained in *1072 the withheld surveys is similar to that already released, and also non-exempt. We therefore reverse the district court’s decision and remand with instructions to enter a finding of segregability. In order to succeed on remand, the Air Force must demonstrate that, unlike the released poll results, the withheld poll results would actually inhibit candor in the decision-making process if made available to the public. See, e.g., Petroleum Info. Corp. v. Dept. of the Interior, 976 F.2d 1429, 1435 (D.C.Cir.1992) (the “key question” is “whether disclosure would tend to diminish candor within [the] agency”); Wolfe, 839 F.2d at 773. See also King v. Dept. of Justice, 830 F.2d 210, 217 (D.C.Cir. 1987) (an agency that chooses to withhold information bears the burden of justifying its decision). If the Air Force proves unable to do so, its withholding decision cannot withstand the requester’s contention that the Air Force’s selection of materials to be released is totally self-serving and unrelated to the purposes of Exemption 5. FOIA was designed to preclude a government agency from cherry-picking the materials to be made public. FOIA operates on the premise that government will function best if its warts as well as its wonders are available for public review.

Reversed and remanded.