JUSTICE SCALIA delivered the opinion of the Court.

Under longstanding precedent of the National Labor Relations Board, an employer who believes that an incumbent union no longer enjoys the support of a majority of its employees has three options: to request a formal, Board-supervised election, to withdraw recognition from the union and refuse to bargain, or to conduct an internal poll of employee support for the union. The Board has held that the latter two are unfair labor practices unless the employer can show that it had a "good-faith reasonable doubt" about the union's majority support. We must decide whether the Board's standard for employer polling is rational and consistent with the National Labor Relations Act, and whether the Board's factual determinations in this case are supported by substantial evidence in the record.

I

Mack Trucks, Inc., had a factory branch in Allentown, Pennsylvania, whose service and parts employees were represented by Local Lodge 724 of the International Association of Machinists and Aerospace Workers, AFL-CIO (Local 724). Mack notified its Allentown managers in May 1990 that it intended to sell the branch, and several of those managers formed Allentown Mack Sales Service, Inc., the petitioner here, which purchased the assets of the business on December 20, 1990, and began to operate it as an independent dealership. From December 21, 1990, to January 1, 1991, Allentown hired 32 of the original 45 Mack employees.

During the period before and immediately after the sale, a number of Mack employees made statements to the prospective owners of Allentown Mack Sales suggesting that the incumbent union had lost support among employees in the bargaining unit. In job interviews, eight employees made statements indicating, or at least arguably indicating, that they personally no longer supported the union. In addition, Ron Mohr, a member of the union's bargaining committee and shop steward for the Mack Trucks service department, told an Allentown manager that it was his feeling that the employees did not want a union, and that "with a new company, if a vote was taken, the Union would lose." 316 N.L.R.B. 1199, 1207 (1995). And Kermit Bloch, who worked for Mack Trucks as a mechanic on the night shift, told a manager that the entire night shift (then five or six employees) did not want the union.

On January 2, 1991, Local 724 asked Allentown Mack Sales to recognize it as the employees' collective-bargaining representative, and to begin negotiations for a contract. The new employer rejected that request by letter dated January 25, claiming a "good faith doubt as to support of the Union among the employees." Id., at 1205. The letter also announced that Allentown had "arranged for an independent poll by secret ballot of its hourly employees to be conducted under guidelines prescribed by the National Labor Relations Board." Ibid. The poll, supervised by a Roman Catholic priest, was conducted on February 8, 1991; the union lost 19 to 13. Shortly thereafter, the union filed an unfair-labor-practice charge with the Board.

The Administrative Law Judge (ALJ) concluded that Allentown was a "successor" employer to Mack Trucks, Inc., and therefore inherited Mack's bargaining obligation and a presumption of continuing majority support for the union. Id., at 1203. The ALJ held that Allentown's poll was conducted in compliance with the procedural standards enunciated by the Board in Struksnes Constr. Co., 165 N.L.R.B. 1062 (1967), but that it violated § 8(a)(1) and 8(a)(5) of the National Labor Relations Act (Act), 49 Stat. 452, as amended, 29 U.S.C. § 158(a)(1) and 158(a)(5), because Allentown did not have an "objective reasonable doubt" about the majority status of the union. The Board adopted the ALJ's findings and agreed with his conclusion that Allentown "had not demonstrated that it harbored a reasonable doubt, based on objective considerations, as to the incumbent Union's continued majority status after the transition." 316 N.L.R.B., at 1199. The Board ordered Allentown to recognize and bargain with Local 724.

On review in the Court of Appeals for the District of Columbia Circuit, Allentown challenged both the facial rationality of the Board's test for employer polling and the Board's application of that standard to the facts of this case. The court enforced the Board's bargaining order, over a vigorous dissent. 83 F.3d 1483 (1996). We granted certiorari. 520 U.S. 1103 (1997).

II

Allentown challenges the Board's decision in this case on several grounds. First, it contends that because the Board's "reasonable doubt" standard for employer polls is the same as its standard for unilateral withdrawal of recognition and for employer initiation of a Board-supervised election (a so-called "Representation Management," or "RM," election), the Board irrationally permits employers to poll only when it would be unnecessary and legally pointless to do so. Second, Allentown argues that the record evidence clearly demonstrates that it had a good-faith reasonable doubt about the union's claim to majority support. Finally, it asserts that the Board has, sub silentio (and presumably in violation of law), abandoned the "reasonable doubt" prong of its polling standard, and recognizes an employer's "reasonable doubt" only if a majority of the unit employees renounce the union. In this Part of our opinion we address the first of these challenges; the other two, which are conceptually intertwined, will be addressed in Parts III and IV.

Courts must defer to the requirements imposed by the Board if they are "rational and consistent with the Act," Fall River Dyeing Finishing Corp. v. NLRB, 482 U.S. 27, 42 (1987), and if the Board's "explication is not inadequate, irrational or arbitrary," NLRB v. Erie Resistor Corp., 373 U.S. 221, 236 (1963). Allentown argues that it is irrational to require the same factual showing to justify a poll as to justify an outright withdrawal of recognition, because that leaves the employer with no legal incentive to poll. Under the Board's framework, the results of a poll can never supply an otherwise lacking "good-faith reasonable doubt" necessary to justify a withdrawal of recognition, since the employer must already have that same reasonable doubt before he is permitted to conduct a poll. Three Courts of Appeals have found that argument persuasive. NLRB v. A.W. Thompson, Inc., 651 F.2d 1141, 1144 (CA5 1981); see also Mingtree Restaurant, Inc. v. NLRB, 736 F.2d 1295 (CA9 1984); Thomas Industries, Inc. v. NLRB, 687 F.2d 863 (CA6 1982).

While the Board's adoption of a unitary standard for polling, RM elections, and withdrawals of recognition is in some respects a puzzling policy, we do not find it so irrational as to be "arbitrary [or] capricious" within the meaning of the Administrative Procedure Act, 5 U.S.C. § 706. The Board believes that employer polling is potentially "disruptive" to established bargaining relationships and "unsettling" to employees, and so has chosen to limit severely the circumstances under which it may be conducted. Texas Petrochemicals Corp., 296 N.L.R.B. 1057, 1061 (1989), enf'd as modified, 923 F.2d 398 (CA5 1991). The unitary standard reflects the Board's apparent conclusion that polling should be tolerated only when the employer might otherwise simply withdraw recognition and refuse to bargain.

It is true enough that this makes polling useless as a means of insulating a contemplated withdrawal of recognition against an unfair-labor-practice charge — but there is more to life (and even to business) than escaping unfair-labor-practice findings. An employer concerned with good employee relations might recognize that abrupt withdrawal of recognition — even from a union that no longer has majority support — will certainly antagonize union supporters, and perhaps even alienate employees who are on the fence. Preceding that action with a careful, unbiased poll can prevent these consequences. The "polls are useless" argument falsely assumes, moreover, that every employer will want to withdraw recognition as soon as he has enough evidence of lack of union support to defend against an unfair-labor-practice charge. It seems to us that an employer whose evidence met the "good-faith reasonable doubt" standard might nonetheless want to withdraw recognition only if he had conclusive evidence that the union in fact lacked majority support, lest he go through the time and expense of an (ultimately victorious) unfair-labor-practice suit for a benefit that will only last until the next election. See Texas Petrochemicals, supra, at 1063. And finally, it is probably the case that, though the standard for conviction of an unfair labor practice with regard to polling is identical to the standard with regard to withdrawal of recognition, the chance that a charge will be filed is significantly less with regard to the polling, particularly if the union wins.

It must be acknowledged that the Board's avowed preference for RM elections over polls fits uncomfortably with its unitary standard; as the Court of Appeals pointed out, that preference should logically produce a more rigorous standard for polling. 83 F.3d, at 1487. But there are other reasons why the standard for polling ought to be less rigorous than the standard for Board elections. For one thing, the consequences of an election are more severe: If the union loses an employer poll it can still request a Board election, but if the union loses a formal election it is barred from seeking another for a year. See 29 U.S.C. § 159(c)(3). If it would be rational for the Board to set the polling standard either higher or lower than the threshold for an RM election, then surely it is not irrational for the Board to split the difference.

III

The Board held Allentown guilty of an unfair labor practice in its conduct of the polling because it "ha[d] not demonstrated that it held a reasonable doubt, based on objective considerations, that the Union continued to enjoy the support of a majority of the bargaining unit employees." 316 N.L.R.B., at 1199. We must decide whether that conclusion is supported by substantial evidence on the record as a whole. Fall River Dyeing, supra, at 42; Universal Camera Corp. v. NLRB, 340 U.S. 474 (1951). Put differently, we must decide whether on this record it would have been possible for a reasonable jury to reach the Board's conclusion. See, e.g., NLRB v. Columbian Enameling Stamping Co., 306 U.S. 292, 300 (1939); Consolidated Edison Co. v. NLRB, 305 U.S. 197, 229 (1938).

Before turning to that issue, we must clear up some semantic confusion. The Board asserted at argument that the word "doubt" may mean either "uncertainty" or "disbelief," and that its polling standard uses the word only in the latter sense. We cannot accept that linguistic revisionism. "Doubt" is precisely that sort of "disbelief" (failure to believe) which consists of an uncertainty rather than a belief in the opposite. If the subject at issue were the existence of God, for example, "doubt" would be the disbelief of the agnostic, not of the atheist. A doubt is an uncertain, tentative, or provisional disbelief. See, e.g., Webster's New International Dictionary 776 (2d ed. 1949) (def. 1: "A fluctuation of mind arising from defect of knowledge or evidence; uncertainty of judgment or mind; unsettled state of opinion concerning the reality of an event, or the truth of an assertion, etc."); 1 The New Shorter Oxford English Dictionary 734 (1993) (def. 1: "Uncertainty as to the truth or reality of something or as to the wisdom of a course of action; occasion or room for uncertainty"); American Heritage Dictionary 555 (3d ed. 1992) (def. 1: "A lack of certainty that often leads to irresolution").

The question presented for review, therefore, is whether, on the evidence presented to the Board, a reasonable jury could have found that Allentown lacked a genuine, reasonable uncertainty about whether Local 724 enjoyed the continuing support of a majority of unit employees. In our view, the answer is no. The Board's finding to the contrary rests on a refusal to credit probative circumstantial evidence, and on evidentiary demands that go beyond the substantive standard the Board purports to apply.

The Board adopted the ALJ's finding that 6 of Allentown's 32 employees had made "statements which could be used as objective considerations supporting a good-faith reasonable doubt as to continued majority status by the Union." 316 N.L.R.B., at 1207. (These included, for example, the statement of Rusty Hoffman that "he did not want to work in a union shop," and "would try to find another job if he had to work with the Union." Id., at 1206.) The Board seemingly also accepted (though this is not essential to our analysis) the ALJ's willingness to assume that the statement of a seventh employee (to the effect that he "did not feel comfortable with the Union and thought it was a waste of $35 a month," i bid.) supported good-faith reasonable doubt of his support for the union — as in our view it unquestionably does. And it presumably accepted the ALJ's assessment that "7 of 32, or roughly 20 percent of the involved employees" was not alone sufficient to create "an objective reasonable doubt of union majority support," id., at 1207. The Board did not specify how many express disavowals would have been enough to establish reasonable doubt, but the number must presumably be less than 16 (half of the bargaining unit), since that would establish reasonable certainty. Still, we would not say that 20% first-hand-confirmed opposition (even with no countering evidence of union support) is alone enough to require a conclusion of reasonable doubt. But there was much more.

For one thing, the ALJ and the Board totally disregarded the effect upon Allentown of the statement of an eighth employee, Dennis Marsh, who said that "he was not being represented for the $35 he was paying." Ibid. The ALJ, whose findings were adopted by the Board, said that this statement "seems more an expression of a desire for better representation than one for no representation at all." Ibid. It seems to us that it is, more accurately, simply an expression of dissatisfaction with the union's performance — which could reflect the speaker's desire that the union represent him more effectively, but could also reflect the speaker's desire to save his $35 and get rid of the union. The statement would assuredly engender an uncertainty whether the speaker supported the union, and so could not be entirely ignored.

But the most significant evidence excluded from consideration by the Board consisted of statements of two employees regarding not merely their own support of the union, but support among the work force in general. Kermit Bloch, who worked on the night shift, told an Allentown manager that "the entire night shift did not want the Union." Ibid. The ALJ refused to credit this, because "Bloch did not testify and thus could not explain how he formed his opinion about the views of his fellow employees." Ibid. Unsubstantiated assertions that other employees do not support the union certainly do not establish the fact of that disfavor with the degree of reliability ordinarily demanded in legal proceedings. But under the Board's enunciated test for polling, it is not the fact of disfavor that is at issue (the poll itself is meant to establish that), but rather the existence of a reasonable uncertainty on the part of the employer regarding that fact. On that issue, absent some reason for the employer to know that Bloch had no basis for his information, or that Bloch was lying, reason demands that the statement be given considerable weight.

Another employee who gave information concerning overall support for the union was Ron Mohr, who told Allentown managers that "if a vote was taken, the Union would lose" and that "it was his feeling that the employees did not want a union." Ibid. The ALJ again objected irrelevantly that "there is no evidence with respect to how he gained this knowledge." Id., at 1208. In addition, the Board held that Allentown "could not legitimately rely on [the statement] as a basis for doubting the Union's majority status," id., at 1200, because Mohr was "referring to Mack's existing employee complement, not to the individuals who were later hired by [Allentown]," ibid. This basis for disregarding Mohr's statements is wholly irrational. Local 724 had never won an election, or even an informal poll, within the actual unit of 32 Allentown employees. Its claim to represent them rested entirely on the Board's presumption that the work force of a successor company has the same disposition regarding the union as did the work force of the predecessor company, if the majority of the new work force came from the old one. See id., at 1197, n. 3; Fall River Dyeing, 482 U.S., at 43, 46-52. The Board cannot rationally adopt that presumption for purposes of imposing the duty to bargain, and adopt precisely the opposite presumption ( i.e., contend that there is no relationship between the sentiments of the two work forces) for purposes of determining what evidence tends to establish a reasonable doubt regarding union support. Such irrationality is impermissible even if, as JUSTICE BREYER, it would further the Board's political objectives.

It must be borne in mind that the issue here is not whether Mohr's statement clearly establishes a majority in opposition to the union, but whether it contributes to a reasonable uncertainty whether a majority in favor of the union existed. We think it surely does. Allentown would reasonably have given great credence to Mohr's assertion of lack of union support, since he was not hostile to the union, and was in a good position to assess antiunion sentiment. Mohr was a union shop steward for the service department, and a member of the union's bargaining committee; according to the ALJ, he "did not indicate personal dissatisfaction with the Union." 316 N.L.R.B., at 1208. It seems to us that Mohr's statement has undeniable and substantial probative value on the issue of "reasonable doubt."

Accepting the Board's apparent (and in our view inescapable) concession that Allentown received reliable information that 7 of the bargaining-unit employees did not support the union, the remaining 25 would have had to support the union by a margin of 17 to 8 — a ratio of more than 2 to 1 — if the union commanded majority support. The statements of Bloch and Mohr would cause anyone to doubt that degree of support, and neither the Board nor the ALJ discussed any evidence that Allentown should have weighed on the other side. The most pro-union statement cited in the ALJ's opinion was Ron Mohr's comment that he personally "could work with or without the Union," and "was there to do his job." Id., at 1207. The Board cannot covertly transform its presumption of continuing majority support into a working assumption that all of a successor's employees support the union until proved otherwise. Giving fair weight to Allentown's circumstantial evidence, we think it quite impossible for a rational factfinder to avoid the conclusion that Allentown had reasonable, good-faith grounds to doubt — to be uncertain about — the union's retention of majority support.

IV

That conclusion would make this a fairly straightforward administrative-law case, except for the contention that the Board's factfinding here was not an aberration. Allentown asserts that, although "the Board continues to cite the words of the good faith doubt branch of its withdrawal of recognition standard," a systematic review of the Board's decisions will reveal that "it has in practice eliminated the good faith doubt branch in favor of a strict head count." Brief for Petitioner 10. The Board denies (not too persuasively) that it has insisted upon a strict head count, but does defend its factfinding in this case by saying that it has regularly rejected similarly persuasive demonstrations of reasonable good-faith doubt in prior decisions. The Court of Appeals in fact accepted that defense, relying on those earlier, similar decisions to conclude that the Board's findings were supported by substantial evidence here. See 83 F.3d, at 1488. That the current decision may conform to a long pattern is also suggested by academic commentary. One scholar, after conducting "[a] thorough review of the withdrawal of recognition case law," concluded:

"Circumstantial evidence, no matter how abundant, is rarely, if ever, enough to satisfy the good-faith doubt test. In practice, the Board deems the test satisfied only if the employer has proven that a majority of the bargaining unit has expressly repudiated the union. Such direct evidence, however, is nearly impossible to gather lawfully. Thus, the Board's good-faith doubt

standard, although ostensibly a highly fact-dependent totality-of-the-circumstances test, approaches a per se rule in application. . . ." Flynn, The Costs and Benefits of "Hiding the Ball": NLRB Policymaking and the Failure of Judicial Review, 75 B.U. L. Rev. 387, 394-395 (1995) (footnotes omitted).

See also Weeks, The Union's Mid-Contract Loss of Majority Support: A Waivering Presumption, 20 Wake Forest L. Rev. 883, 889 (1984). Members of this Court have observed the same phenomenon. See NLRB v. Curtin Matheson Scientific, Inc., 494 U.S. 775, 797 (1990) (REHNQUIST, C.J., concurring) ("[S]ome recent decisions suggest that [the Board] now requires an employer to show that individual employees have `expressed desires' to repudiate the incumbent union in order to establish a reasonable doubt of the union's majority status"); id., at 799 (Blackmun, J., dissenting) ("[T]he Board appears to require that good-faith doubt be established by express avowals of individual employees").

It is certainly conceivable that an adjudicating agency might consistently require a particular substantive standard to be established by a quantity or character of evidence so far beyond what reason and logic would require as to make it apparent that the announced standard is not really the effective one. And it is conceivable that in certain categories of cases an adjudicating agency which purports to be applying a preponderance standard of proof might so consistently demand in fact more than a preponderance, that all should be on notice from its case law that the genuine burden of proof is more than a preponderance. The question arises, then, whether, if that should be the situation that obtains here, we ought to measure the evidentiary support for the Board's decision against the standards consistently applied rather than the standards recited. As a theoretical matter (and leaving aside the question of legal authority), the Board could certainly have raised the bar for employer polling or withdrawal of recognition by imposing a more stringent requirement than the reasonable-doubt test, or by adopting a formal requirement that employers establish their reasonable doubt by more than a preponderance of the evidence. Would it make any difference if the Board achieved precisely the same result by formally leaving in place the reasonable-doubt and preponderance standards, but consistently applying them as though they meant something other than what they say? We think it would.

The Administrative Procedure Act, which governs the proceedings of administrative agencies and related judicial review, establishes a scheme of "reasoned decisionmaking." Motor Vehicle Mfrs. Assn. of United States, Inc. v. State Farm Mut. Automobile Ins. Co., 463 U.S. 29, 52 (1983). Not only must an agency's decreed result be within the scope of its lawful authority, but the process by which it reaches that result must be logical and rational. Courts enforce this principle with regularity when they set aside agency regulations which, though well within the agencies' scope of authority, are not supported by the reasons that the agencies adduce. See SEC v. Chenery Corp., 318 U.S. 80 (1943); SEC v. Chenery Corp., 332 U.S. 194 (1947). The National Labor Relations Board, uniquely among major federal administrative agencies, has chosen to promulgate virtually all the legal rules in its field through adjudication rather than rulemaking. See, e.g., NLRB v. Bell Aerospace Co., 416 U.S. 267, 294-295 (1974). (To our knowledge, only one regulation has ever been adopted by the Board, dealing with the appropriate size of bargaining units in the health care industry. See 29 C.F.R. § 103.30 (1997).) But adjudication is subject to the requirement of reasoned decisionmaking as well. It is hard to imagine a more violent breach of that requirement than applying a rule of primary conduct or a standard of proof which is in fact different from the rule or standard formally announced. And the consistent repetition of that breach can hardly mend it.

Reasoned decisionmaking, in which the rule announced is the rule applied, promotes sound results, and unreasoned decisionmaking the opposite. The evil of a decision that applies a standard other than the one it enunciates spreads in both directions, preventing both consistent application of the law by subordinate agency personnel (notably ALJ's), and effective review of the law by the courts. These consequences are well exemplified by a recent withdrawal-of-recognition case in which the Board explicitly reaffirmed its adherence to the preponderance-of-the-evidence standard. One of the Board's ALJ's, interpreting the agency's prior cases as many others have, had concluded that the Board in fact required "`clear, cogent, and convincing'" evidence that the union no longer commanded a majority. Laidlaw Waste Systems, Inc., 307 N.L.R.B. 1211 (1992). On review the Board rejected that standard, insisting that "in order to rebut the presumption of an incumbent union's majority status, an employer must show by a preponderance of the evidence . . . objective factors sufficient to support a reasonable and good-faith doubt of the union's majority." Ibid. So far, so good. The Board then went on to add, however, that "[t]his is not to say that the terms `clear, cogent, and convincing' have no significance at all in withdrawal of recognition cases." Ibid. It then proceeded to make the waters impenetrably muddy with the following:

"It is fair to say that the Board will not find that an employer has supported its defense by a preponderance of the evidence if the employee statements and conduct relied on are not clear and cogent rejections of the union as a bargaining agent, i.e., are simply not convincing manifestations, taken as a whole, of a loss of majority support. The opposite of `clear, cogent, and convincing' evidence in this regard might be fairly described as `speculative, conjectural, and vague' — evidence that plainly does not meet the preponderance-of-the-evidence burden of proof." Id., at 1211-1212.

Each sentence of this explanation is nonsense, and the two sentences together are not even compatibly non-sensical. "Preponderance of the evidence" and "clear and convincing evidence" describe well known, contrasting standards of proof. To say, as the first sentence does, that a preponderance standard demands "clear and convincing manifestations, taken as a whole" is to convert that standard into a higher one; and to say, as the second sentence does, that whatever is not "speculative, conjectural, and vague" meets the "clear, cogent, and convincing" standard is to reconvert that standard into a lower one. And the offsetting errors do not produce rationality but compounded confusion. If the Board's application of the preponderance standard is indeed accurately described by this passage, it is hard for the ALJ to know what to do with the next case.

A case like Laidlaw, or a series of cases that exemplify in practice its divorcing of the rule announced from the rule applied, also frustrates judicial review. If revision of the Board's standard of proof can be achieved thus subtly and obliquely, it becomes a much more complicated enterprise for a court of appeals to determine whether substantial evidence supports the conclusion that the required standard has or has not been met. It also becomes difficult for this Court to know, when certiorari is sought, whether the case involves the generally applicable issue of the Board's adoption of an unusually high standard of proof, or rather just the issue of an allegedly mistaken evidentiary judgment in the particular case. An agency should not be able to impede judicial review, and indeed even political oversight, by disguising its policymaking as factfinding.

Because reasoned decisionmaking demands it, and because the systemic consequences of any other approach are unacceptable, the Board must be required to apply in fact the clearly understood legal standards that it enunciates in principle, such as good-faith reasonable doubt and preponderance of the evidence. Reviewing courts are entitled to take those standards to mean what they say, and to conduct substantial-evidence review on that basis. Even the most consistent and hence predictable Board departure from proper application of those standards will not alter the legal rule by which the agency's factfinding is to be judged.

That principle is not, as JUSTICE BREYER suggests, inconsistent with our decisions according "substantial deference to an agency's interpretation of its own regulations." Thomas Jefferson Univ. v. Shalala, 512 U.S. 504, 512 (1994). Substantive review of an agency's interpretation of its regulations is governed only by that general provision of the Administrative Procedure Act which requires courts to set aside agency action that is "arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law," 5 U.S.C. § 706(2)(A). It falls well within this text to give the agency the benefit of the doubt as to the meaning of its regulation. On-the-record agency factfinding, however, is also governed by a provision that requires the agency action to be set aside if it is "unsupported by substantial evidence," § 706(2)(E) — which is the very specific requirement at issue here. See also 29 U.S.C. § 160(e) ("The findings of the Board with respect to questions of fact if supported by substantial evidence on the record considered as a whole shall be conclusive"). The "substantial evidence" test itself already gives the agency the benefit of the doubt, since it requires not the degree of evidence which satisfies the court that the requisite fact exists, but merely the degree which could satisfy a reasonable factfinder. See Columbian Enameling Stamping Co., 306 U.S., at 300. This is an objective test, and there is no room within it for deference to an agency's eccentric view of what a reasonable factfinder ought to demand. We do not, moreover (we could not possibly), search to find revisions of the agency's rules — revisions of the requisite fact that the adjudication is supposed to determine — hidden in the agency's factual findings. In the regime envisioned by JUSTICE BREYER — a regime in which inadequate factual findings become simply a revision of the standard that the Board's (adjudicatorily adopted) rules set forth, thereby converting those findings into rule interpretations to which judges must defer — the "substantial evidence" factual review provision of the Administrative Procedure Act becomes a nullity.

The Board can, of course, forthrightly and explicitly adopt counter-factual evidentiary presumptions (which are in effect substantive rules of law) as a way of furthering particular legal or policy goals — for example, the Board's irrebutetable presumption of majority support for the union during the year following certification, see, e.g., Station KKHI, 284 N.L.R.B. 1339, 1340 (1987), enf'd, 891 F.2d 230 (CA9 1989). The Board might also be justified in forthrightly and explicitly adopting a rule of evidence that categorically excludes certain testimony on policy grounds, without reference to its inherent probative value. (Such clearly announced rules of law or of evidentiary exclusion would of course be subject to judicial review for their reasonableness and their compatibility with the Act.) That is not the sort of Board action at issue here, however, but rather the Board's allegedly systematic undervaluation of certain evidence, or allegedly systematic exaggeration of what the evidence must prove. See, e.g., Westbrook Bowl, 293 N.L.R.B. 1000, 1001, n. 11 (1989) ("The Board has stated that `testimony concerning conversations directly with the employees involved . . . is much more reliable than testimony concerning merely a few employees ostensibly conveying the sentiments of their fellows'"), quoting Sofco, Inc., 268 N.L.R.B. 159, 160, n. 10 (1983). When the Board purports to be engaged in simple factfinding, unconstrained by substantive presumptions or evidentiary rules of exclusion, it is not free to prescribe what inferences from the evidence it will accept and reject, but must draw all those inferences that the evidence fairly demands. "Substantial evidence" review exists precisely to ensure that the Board achieves minimal compliance with this obligation, which is the foundation of all honest and legitimate adjudication.

For the foregoing reasons, we need not determine whether the Board has consistently rejected or discounted probative evidence so as to cause "good-faith reasonable doubt" or "preponderance of the evidence" to mean something more than what the terms connote. The line of precedents relied on by the ALJ and the Court of Appeals could not render irrelevant to the Board's decision, and hence to our review, any evidence that tends to establish the existence of a good-faith reasonable doubt. It was therefore error, for example, for the ALJ to discount Ron Mohr's opinion about lack of union support because of "the Board's historical treatment of unverified assertions by an employee about another employee's sentiments." 316 N.L.R.B., at 1208. And it was error for the Court of Appeals to rely upon the fact that "[t]he Board has consistently questioned the reliability of reports by one employee of the antipathy of other employees toward their union." 83 F.3d, at 1488, citing Westbrook Bowl, supra, at 1001, n. 11; Sofco, Inc., supra, at 160, n. 10. Assuming that those assessments of the Board's prior behavior are true, they nonetheless provide no justification for the Board's factual inferences here. Of course, the Board is entitled to be skeptical about the employer's claimed reliance on secondhand reports when the reporter has little basis for knowledge, or has some incentive to mislead. But that is a matter of logic and sound inference from all the circumstances, not an arbitrary rule of disregard to be extracted from prior Board decisions.

The same is true of the Board precedents holding that "an employee's statements of dissatisfaction with the quality of union representation may not be treated as opposition to union representation," and that "an employer may not rely on an employee's anti-union sentiments, expressed during a job interview in which the employer has indicated that there will be no union." 83 F.3d, at 1488, citing Destileria Serralles, Inc., 289 N.L.R.B. 51 (1988), enf'd, 882 F.2d 19 (CA1 1989), and Middleboro Fire Apparatus, Inc., 234 N.L.R.B. 888, 894, enf'd, 590 F.2d 4 (CA1 1978). It is of course true that such statements are not clear evidence of an employee's opinion about the union — and if the Board's substantive standard required clear proof of employee disaffection, it might be proper to ignore such statements altogether. But that is not the standard, and, depending on the circumstances, the statements can unquestionably be probative to some degree of the employer's good-faith reasonable doubt.

* * *

We conclude that the Board's "reasonable doubt" test for employer polls is facially rational and consistent with the Act. But the Board's factual finding that Allentown Mack Sales lacked such a doubt is not supported by substantial evidence on the record as a whole. The judgment of the Court of Appeals for the Distrcit of Columbia Circuit is therefore reversed, and the case is remanded with instructions to deny enforcement.

It is so ordered.