Roy Arthur Nelson and Frank Brew-ton were indicted for robbery of a federally-insured institution in violation of 18 U.S.C. § 2113 (a) (1964). Brewton was found incompetent to stand trial. Nelson was tried separately and convicted. He has appealed on three grounds, all of which relate to the use of circumstantial evidence to secure his conviction.

The government offered direct evidence of the following facts. Brewton entered a bank and presented a teller with a written demand for money. The teller handed Brewton $627 in currency, including five marked $20 bills. Meanwhile, an unidentified person was observed sitting in a car in an adjacent parking lot, racing the engine. Brewton fled from the bank to the waiting car and entered on the passenger side. The car immediately sped away. Shortly thereafter, a police officer, alerted to these incidents, observed the car, with two male occupants, at an intersection some blocks away. The car fled. The officer pursued at high speed. After a chase the car slowed down, defendant alighted from the driver’s side and ran, and was captured. Currency in the amount of $125 was taken from his person. The car, driverless, crashed into a tree. Brewton emerged from the wreck, and was arrested after attempting to conceal $502 in currency, including the marked bills taken from the bank, under an adjacent building. Ten to fifteen minutes elapsed between the robbery of the bank and defendant’s flight from the car.

Defendant asserts that since he was charged as a principal in the bank robbery rather than as an accessory after the fact, the government was required to prove that he had actual knowledge that Brewton intended to rob the bank. We assume, arguendo, that proof of precisely that specific knowledge was required.

Defendant contended below, and in this court, that such proof was lacking. He argued that if such knowledge could be inferred at all, the inference must be based upon the prior inference that he was the man waiting in the car while the robbery occurred — and such an “inference upon an inference” was precluded by law. Further, he argued that even if the jury were permitted to infer that he knowingly acted as the “get-away” driv er, there was no evidence that he knew Brewton planned to commit a robbery, as distinguished from some other illegal act, or planned to rob the bank, and not one of the several stores and offices in the area, and that circumstantial evidence which does not “exclude every hypothesis but that of guilt” is insufficient as a matter of law.

The court denied a motion for acquittal based on these grounds, and rejected proposed instructions embodying the theories that a conviction could not be based upon inferences drawn from other inferences, or upon circumstances “which while consistent with guilt, are not inconsistent with innocence.”

The legal theories upon which defendant relies, although clearly wrong, are repeatedly asserted in the trial courts of this circuit and in fruitless appeals to this court. It would be a boon to both the parties and the courts if they could be laid finally to rest.

I

For at least a third of a century this court has rejected the notion that it is improper to infer a fact at issue from other facts which have been established by circumstantial evidence. E. K. Wood Lumber Co. v. Andersen, 81 F.2d 161, 166 (9th Cir. 1936); Boss v. United States, 103 F.2d 600, 606 (9th Cir. 1939); Fegles Construction Co. v. McLaughlin Construction Co., 205 F.2d 637, 639-640 (9th Cir. 1953); Toliver v. United States, 224 F.2d 742, 745 (9th Cir. 1955); Medrano v. United States, 315 F.2d 361, 362 (9th Cir. 1963); De-vore v. United States, 368 F.2d 396, 399 (9th Cir. 1966). As Professor Wig-more has said, “[tjhere is no such orthodox rule; nor can be. If there were, hardly a single trial could be adequately presented.” 1 Wigmore, Evidence, § 41, at 435 (3rd ed. 1940).

The error in this discredited doctrine is clearly reflected in the defendant’s formulation: it assumes that a fact established by circumstantial evidence is not a “proven fact.” But as we have repeatedly said, circumstantial evidence is not inherently less probative than direct evidence. Under some conditions it may even be more reliable, as this case illustrates.

The intermediate fact at issue here was whether defendant was the driver of the car waiting in the parking lot. That fact was established to a moral certainty by circumstances proven by un-contradicted and unquestioned testimony. Unless defendant was Brewton’s accomplice waiting in the get-away car, it is all but inconceivable that he would have been driving that car with Brewton as a passenger a few minutes after Brewton ran from the bank to the car and was driven from the scene; that he would have had part of the stolen currency in his possession, and Brewton the rest; and that upon seeing the police officer he would have driven away at high speed, and later fled from the office on foot.

If none of this circumstantial evidence had been available and the only evidence offered had been a courtroom identification of the defendant by a witness who had a fleeting glimpse of the driver as the car stood in the parking lot, the truth of the fact that defendant was that man would not have been established with equal certainty.

Of course either direct or circumstantial evidence may fail to prove the fact in issue — direct evidence because the credibility of the witness is destroyed; circumstantial evidence for that reason, or because the inference from the proven circumstances to the fact in issue is too speculative, or remote. Whether such a failure has occurred is an appropriate inquiry in any case — be the evidence direct, circumstantial, or both. But since under some conditions circumstantial evidence may be equally or more reliable than direct evidence, it would be wholly irrational to impose an absolute bar upon the use of circumstantial evidence to prove any fact, including a fact from which another fact is to be inferred.

The trial court therefore properly refused to instruct the jury that “one inference may not be based upon another inference to support a conclusion of fact.” It would be error for the jury, the trial court, or this court, to apply such an arbitrary formula in the performance of their respective roles in the fact-finding process.

II

It is also clear that the court properly rejected defendant’s proposed instruction embodying a variation on the theme that circumstantial evidence must exclude every hypothesis but that of guilt.

This much, at least, is settled by Holland v. United States, 348 U.S. 121 , 75 5. Ct. 127, 99 L.Ed. 150 (1955). What the Holland court said is brief, and well worth repeating in full:

The Supreme Court did more than reject the particular instruction before it; it clearly stated that no instruction is to be given distinguishing the manner in which direct and circumstantial evidence are to be weighed. Since circumstantial and testimonial evidence are indistinguishable so far as the jury’s fact-finding function is concerned, all that is to be required of the jury is that it weigh all of the evidence, direct or circumstantial, against the standard of reasonable doubt.

Our holdings are in accord with this interpretation of Holland. Mull v. United States, 402 F.2d 571, 575 (9th Cir. 1968); Ramirez v. United States, 350 F.2d 306, 307-308 (9th Cir. 1965); Armstrong v. United States, 327 F.2d 189, 194 (9th Cir. 1964); Strangway v. United States, 312 F.2d 283, 285 (9th Cir. 1963). Indeed, this was the rule in this circuit prior to Holland. Samuel v. United States, 169 F.2d 787, 791 (9th Cir. 1948); McCoy v. United States, 169 F.2d 776, 784 (9th Cir. 1948); see also Penosi v. United States, 206 F.2d 529, 530 (9th Cir. 1953).

Thus the trial court properly instructed the jury that “the law makes no distinction between direct and circumstantial evidence but simply requires that the jury be satisfied of the defendant’s guilt beyond a reasonable doubt from all the evidence in the case,” including “such reasonable inferences as seem justified, in the light of your own experience.” Under Holland and the uniform decisions of this court, no additional instruction was required, and none would have been proper. The contrary dictum in Matthews v. United States, 394 F.2d 104, 105 (9th Cir. 1968), must therefore be disregarded.

Ill

One question remains: Did the district court properly overrule the motion for acquittal?

Defendant contends that whether or not he was entitled to the proposed jury instruction, this court, in reviewing the sufficiency of the evidence, must ask “whether reasonable minds could find that the evidence excludes every hypothesis but that of guilt.”

The test to be applied by the trial court in deciding a motion for acquittal in a criminal case, and the test to be applied by this court in reviewing that decision, are, as a practical matter, identical. 8 J. Moore, Federal Practice, ¶ 29.06, at 29-19 (1968). It is also undisputed that the evidence must be taken in the light most favorable to the verdict • — that it is the exlusive function of the jury to determine the credibility of witnesses, resolve evidentiary conflicts, and draw reasonable inferences from proven facts. Therefore the reviewing court must assume that the jury resolved all such matters in a manner which would support the verdict. 4 Barron, Federal Practice and Procedure, § 2221, at 235 (1951). Beyond this, however, the substance of the standard to be applied is subject to much confusion and dispute.

The opinions of our court dealing with this issue combine various threads of doctrine in such a wide variety of ways that no simple analysis can accurately reflect their content. Nonetheless, it is convenient to deal with the issues as involving two different, though overlapping, questions.

The first is whether the test of the sufficiency of the evidence in a criminal case is the same as that applied in civil cases — namely, whether there is substantial evidence to support a judgment adverse to the moving party- — or whether a different test is to applied in criminal cases reflecting the burden of establishing guilt beyond a reasonable doubt.

Some courts, led by the Court of Appeals for the Second Circuit, apply the substantial evidence test in criminal as well as civil cases. There is language in Supreme Court opinions, and in opinions of this court, supporting this approach.

However, other courts — indeed, most courts — “have included the standard of persuasion as part of their criterion of sufficiency.” A. Goldstein, The State and the Accused: Balance of Advantage in Criminal Procedure, 69 Yale L.J. 1149 , 1157 (1960). Thus, in American Tobacco Co. v. United States, 328 U.S. 781 , 787 n. 4, 66 S.Ct. 1125 , 1128, 90 L.Ed. 1575 (1946), the Supreme Court stated: “The verdict in a criminal case is sustained only when there is ‘relevant evidence from which the jury could properly find or infer, beyond a reasonable doubt, that the accused is guilty. Mortenson v. United States, 322 U.S. 369, 374 , 64 S.Ct. 1037, 1040 , 88 L.Ed. 1331 .” (Emphasis added.) Similar expressions appear in an overwhelming majority of our opinions reviewing the sufficiency of the evidence in criminal cases.

The function which the court performs in passing upon a motion for acquittal suggests that this latter view is correct. That function is to determine whether the jury in arriving at the verdict has acted within the bounds of its authority. Since it is the jury’s duty to acquit unless guilt is established beyond a reasonable doubt, the reviewing court may properly inquire whether the evidence, considered most favorably to the government, was such as to permit a rational conclusion by the jury that the accused was guilty beyond a reasonable doubt.

The second question is that specifically raised by the defendant, namely, whether the court is also to inquire whether the evidence “excludes every hypothesis but that of guilt.”

Precisely as put, the test is unquestionably wrong. Although we have frequently stated the rule as defendant does, it has never been held that the evidence must exclude “every hypothesis,” as distinguished from every reasonable hypothesis, of innocence. Furthermore, in applying the test, the question is not whether the court itself would find that every reasonable hypothesis of innocence had been excluded, but rather whether the jurors could reasonably arrive at that conclusion.

Thus, as first stated in this court in Stoppelli v. United States, 183 F.2d 391, 393 (9th Cir. 1950), the test was as follows:

This formulation, commonly condensed to “whether ‘reasonable minds could find the evidence excludes every hypothesis but that of guilt,’ ” appears frequently in our opinions.

Yet this is precisely the standard which was rejected by the Supreme Court in Holland v. United States, supra, 348 U.S. at 139 , 75 S.Ct. 127 , as a guide for the jury, on the ground that it was “confusing and incorrect.” Id. at 140 , 75 S.Ct. 127 . Our opinions demonstrate that it is equally confusing as a guide for the reviewing court. Moreover, if it is “incorrect” as an instruction defining the jury’s duty, it must be equally “incorrect” as a test for determining whether the jury has performed its duty within the limits fixed by the instructions. Accordingly, most courts have held that its use as a test of the sufficiency of the evidence on review is inconsistent with Holland.

The “reasonable hypothesis” test was formulated for the evaluation of circumstantial evidence; it is often referred to as the “circumstantial evidence test.” See, e. g., Comment, Sufficiency of Circumstantial Evidence in a Criminal Case, 55 Colum.L.Rev. 549 (1955). As we have noted, the Supreme Court rejected the test in Holland on the premise that there is no essential difference in the mental processes required of the jury in weighing direct and circumstantial evidence. As to both, “the jury must use its experience with people and events in weighing the probabilities. If the jury is convinced beyond a' reasonable doubt, we can require no more.” Holland, 348 U.S. at 140 , 75 S.Ct. at 138 .

The key word is “probabilities.” The jury cannot determine that a proposition is true or false, but only that it is more or less probable. Guilt “is proved beyond a reasonable doubt if it proved not only to be more probable than its contradictory but much more probable than its contradictory.” Adler & Michael, The Trial of an Issue of Fact I, 34 Colum.L.Rev. 1224, 1256 (1934). The required degree of probability is reached if the jury is free of “the kind of doubt that would make a person hesitate to act” in the more serious and important affairs of his own life. Holland v. United States, supra, 348 U.S. at 140 , 75 S.Ct. at 138 .

It adds only an illusion of certainty, and is both misleading and wrong, to attempt to describe this broad exercise of practical judgment in abstract generalizations borrowed from the terminology of formal logic.

The “reasonable hypothesis” test does not reflect what juries and reviewing courts in reality do. Juries constantly convict, and the convictions are duly affirmed, on evidence upon which none would hesitate to act but which cannot be said to exclude as a matter of inexorable logic, every reasonable hypothetical consistent with innocence.

Moreover, the impression left by appellate court opinions is that the “reasonable hypothesis” standard may lead to serious departures from the proper appellate role in evaluating the sufficiency of evidence. Courts following the rule exhibit a noticeable tendency to divide the evidence into separate lines.of proof, and analyze and test each line of proof independently of others rather than considering the evidence as an interrelated whole. The sufficiency of the evidence is often tested against theoretical and speculative possibilities not fairly raised by the record, and inferences are sometimes considered which, though entirely possible or even probable, are drawn from evidence which the jury may have disbelieved.

We affirm the denial of the motion for acquittal in this case because we are satisfied that the jurors reasonably could decide that they would not hesitate to act in their own serious affairs upon factual assumptions as probable as the conclusion that defendant planned and executed the robbery of the bank as a joint venture with Brewton in which each carried out a prearranged role.

Affirmed.

. The following instruction was requested and refused: “Permissible inferences from the evidence must be based upon proven facts and one inference may not be based upon another inference to support a conclusion of fact.” . United States v. Rux, 412 F.2d 331 (9th Cir. 1969); Hiram v. United States, 354 F.2d 4, 6 (9th Cir. 1965); Medrano v. United States, 315 F.2d 361, 362 (9th Cir. 1963). See also Holland v. United States, 348 U.S. 121, 140 , 75 S.Ct. 127 , 99 L.Ed. 150 (1954). . 8 J. Moore, Federal Practice, If 29.06, at 29-21, 29-22 (1968); Osborne, The Problem of Proof, 216, 220 (2d ed. 1926); 1 Wigmore, Evidence, § 26, at 403 (3d ed. 1940); Burns, Weighing Circumstantial Evidence, 2 S.Dak.L.Rev. 36 (1957); Comment, Sufficiency of Circumstantial Evidence in a Criminal Case, 55 Colum.L.Rev. 549, 557-58 (1955). . As the Supreme Court pointed out in United States v. Wade, 388 U.S. 218 , 87 S.Ct. 1926 , 18 L.Ed.2d 1149 (1967), “the annals of criminal law are rife with instances of mistaken [eyewitness] identifications.” 388 U.S. at 228 , 87 S. Ct. at 1933 . See also Simmons v. United States, 390 U.S. 377, 383-384 , 88 S.Ct. 967 , 19 L.Ed.2d 1247 (1968). . 1 Wigmore, supra, § 41, at 435; Comment, 55 Colum.L.Rev., supra, at 557. Compare Direct Sales Co. v. United States, 319 U.S. 703 , 63 S.Ct. 1265 , 87 L.Ed. 1674 (1943), with Ingram v. United States, 360 U.S. 672 , 79 S.Ct. 1314 , 3 L.Ed.2d 1503 (1959). . Defendant proposed the following instruction : “You may not base a finding of the existence of any fact in issue upon the basis of circumstantial evidence if the circumstances relied upon are consistent with another conclusion, even if they are consistent with a conclusion of the truth of the fact in issue. Proof of circumstances which while consistent with guilt, are not inconsistent with innocence, will not support a conviction.” . Neither of the eases cited in support of the dictum in Matthews, Davis v. United States, 382 F.2d 221 (9th Cir. 1967), and Whaley v. United States, 362 F.2d 938 (9th Cir. 1966), involved the propriety of jury instructions. . Authorities are collected on the basis of a somewhat different analysis in Castro v. United States, 323 F.2d 683 n. 1 & 2 (9th Cir. 1963) and accompanying text. The Castro analysis follows that utilized by several commentators. See A. Goldstein, The State and the Accused: Balance of Advantage in Criminal Procedure, 69 Yale L.J. 1149 , 1159 (1960); Note, The Motion for Acquittal: A Neglected Safeguard, 70 Yale L.J. 1151 , 1153 (1961); Comment, 55 Colum.L. Rev., supra, at 549-50. . 8 J. Moore, Federal Practice, ¶ 29.06, at 29-24 (1968); A. Goldstein, The State and the Accused, supra, 69 Yale L.J. at 1159. . Woodby v. Immigration and Naturalization Serv., 385 U.S. 276, 282 , 87 S.Ct. 483 , 17 L.Ed.2d 362 (1966); Glasser v. United States, 315 U.S. 60, 80 , 62 S.Ct. 457 , 86 L.Ed. 680 (1942); United States v. Socony-Vacuum Oil Co., 310 U.S. 150, 254 , 60 S.Ct. 811 , 84 L.Ed. 1129 (1940); Pierce v. United States, 252 U.S. 239, 251-252 , 40 S.Ct. 205 , 64 L.Ed. 542 (1920); Abrams v. United States, 250 U.S. 616, 619 , 40 S.Ct. 17 , 63 L.Ed. 1173 (1919); Stilson v. United States, 250 U.S. 583, 588 , 40 S.Ct. 28 , 63 L.Ed. 1154 (1919). . E. g., Peek v. United States, 321 F.2d 934, 945 (9th Cir. 1963); Foster v. United States, 318 F.2d 684, 689-690 (9th Cir. 1963); Miller v. United States, 302 F.2d 659, 660 (9th Cir. 1962); Bowler v. United States, 249 F.2d 806, 807 (9th Cir. 1957); Randall v. United States, 215 F.2d 587, 589 , 15 Alaska 135 (9th Cir. 1954). . Curley v. United States, 81 U.S.App. D.C. 389, 160 F.2d 229, 232 (1947); 8 J. Moore, supra, ¶ 29.02, at 29-5, 29-6; A. Goldstein, The State and the Accused, supra, 69 Yale L.J. at 1155. . E. g., United States v. Rux, 412 F.2d 331 (9th Cir. 1969); Miller v. United States, 382 F.2d 583, 587 (9th Cir. 1967); Davis v. United States, 382 F.2d 221, 224 (9th Cir. 1967); Urquidi v. United States, 371 F.2d 654, 657 (9th Cir. 1967); Whaley v. United States, 362 F.2d 938, 939 (9th Cir. 1966); Kaplan v. United States, 329 F.2d 561, 563 (9th Cir. 1964). . Obviously the evidence need not exclude every speculative or fanciful hypothesis of innocence. Urquidi v. United States, 371 F.2d 654, 657 (9th Cir. 1967); Hiram v. United States, 354 F.2d 4, 6 (9th Cir. 1965); Bowler v. United States, 249 F.2d 806, 810 (9th Cir. 1957); Stoppelli v. United States, 183 F. 2d 391, 393-394 (9th Cir. 1950); McCoy v. United States, 169 F.2d 776, 785-786 (9th Cir. 1948). See also Fegles Constr. Co. v. McLaughlin Constr. Co., 205 F.2d 637, 639 (9th Cir. 1953). . E. g., Kotteakos v. United States, 328 U.S. 750, 763-764 , 66 S.Ct. 1239 , 90 L. Ed. 1557 (1946); Stilson v. United States, 250 U.S. 583, 588 , 40 S.Ct. 28 , 63 L.Ed. 1154 (1919); Pierce v. United States, 252 U.S. 239, 251-252 , 40 S.Ct. 205 , 64 L.Ed. 542 (1920); Crawford v. United States, 126 U.S.App.D.C. 156 , 375 F.2d 332, 334 (1967); Figueroa v. United States, 352 F.2d 587, 592 (9th Cir. 1965); Barnard v. United States, 342 F.2d 309, 317 (9th Cir. 1965); Peek v. United States, 321 F.2d 934, 945 (9th Cir. 1963); Strauss v. United States, 311 F.2d 926, 931 (5th Cir. 1963); Remmer v. United States, 205 F.2d 277, 287-288 (9th Cir. 1953); Curley v. United States, 81 U.S.App.D.C. 389 , 160 F.2d 229, 233 (1947) ; See also Sentilles v. Inter-Carribbean Shipping Corp., 361 U.S. 107, 110 , 80 S.Ct. 173 , 4 L.Ed.2d 142 (1959); Lavender v. Kurn, 327 U.S. 645, 652 , 66 S.Ct. 740 , 90 L.Ed. 916 (1946); Tennant v. Peoria & P. U. Ry., 321 U.S. 29, 35 , 64 S.Ct. 409 , 88 L.Ed. 520 (1944). . Byrnes v. United States, 327 F.2d 825 , 829 n.5a (9th Cir. 1964); Bolen v. United States, 303 F.2d 870, 874 (9th Cir. 1962); Cape v. United States, 283 F.2d 430, 433 (9th Cir. 1980); Elwert v. United States, 231 F.2d 928, 933, 935 (9th Cir. 1956); Schino v. United States, 209 F.2d 67, 72 (9th Cir. 1953); Remmer v. United States, 205 F.2d 277, 287-288 (9th Cir. 1953). . See notes 13 and 19. In Mull v. United States, 402 F.2d 571, 575 (9th Cir. 1968), and Armstrong v. United States, 327 F.2d 189, 194-195 (9th Cir. 1964), we held that it was not error to refuse to instruct the jury that where the evidence was susceptible of two reasonable constructions, one pointing to guilt, the other to innocence, the jury must adopt the latter. But in Miller v. United States, 382 F.2d 583, 587 (9th Cir. 1967), we reversed a conviction because, in our view, two equally valid inferences from the evidence were possible and the jury failed to draw the inference favoring the defendant. . E. g., Dirring v. United States, 328 F.2d 512, 515 (1st Cir. 1985); United States v. Press, 336 F.2d 1003, 1010 (2d Cir. 1964) ; United States v. Tutino, 269 F.2d 488, 490 (2d Cir. 1959) ; United States v. Allard, 240 F.2d 840, 841 (3d Cir. 1957) ; White v. United States, 279 F.2d 740, 748 (4th Cir. 1960); United States v. Comer, 288 F.2d 174, 175 (6th Cir. 1961); Corbin v. United States, 253 F.2d 646, 648-649 (10th Cir. 1958). See also 8 J. Moore, Federal Practice, ¶ 29.06, at 29-21 n.6 (1958) ; Comment, 55 Colum.L.Rev. supra, at 558 (1955). The Fifth Circuit takes a contrary view. E. g., Riggs v. United States, 280 F.2d 949, 955 (5th Cir. 1960); Cuthbert v. United States, 278 F.2d 220, 224-225 (5th Cir. 1960). It is not readily apparent why our court failed to adopt this position at the outset. In two post-Holland cases, the court reasserted the pre-Holland “reasonable hypothesis” test as the standard to be applied on motion for acquittal; but Holland was not cited, and, so far as appears, was not called to the court’s attention. Cape v. United States, 283 F.2d 430, 433 (9th Cir. 1960); Bolen v. United States, 303 F.2d 870, 874 (9th Cir. 1962). Later, in Strangway v. United States, 312 F.2d 283, 285 (9th Cir. 1963), the court recognized that Holland had rejected the “reasonable hypothesis” rule as a jury instruction. In the following year, the test to be applied in reviewing the denial of a motion for acquittal was again before the court in Byrnes v. United States, 327 F.2d 825 (9th Cir. 1964). The court stated: “Appellant urges us to accept his argument that in order to convict, inferences that may be drawn from circumstantial evidence must not only be consistent with a defendant’s guilt, but inconsistent with every reasonable hypothesis of innocence * * *. This is the old rule which the Supreme Court disapproved in Holland v. United States, 1954, 348 U.S. 121 , 75 S.Ct. 127 , 99 L.Ed. 150 * * *.” 327 F.2d at 829 n.5a. Thus we assumed that Holland had disapproved the “reasonable hypothesis” rule not only as a jury instruction but also as a test of the sufficiency of the evidence. The Byrnes court then went on to say, however, without explanation or citation of authority, that “[t]he current test is whether ‘reasonable minds could find that the evidence excludes every hypothesis but that of guilt.’ ” Id. This would ap.pear to say that Holland modified the “reasonable hypothesis” test only by making the reference what “reasonable minds could find.” But as noted in the text, Holland rejected the “reasonable hypothesis” test entirely. The formulation which the court quotes in Byrnes as the “current test” is not a Holland innovation at all. Indeed, except for the omission of the word “reasonable” before the phrase “every hypothesis,” the quotation appears to be taken from Remmer v. United States, 205 F.2d 277, 287-288 (9th Cir. 1953), which was decided a year prior to Holland and stated the test as it had existed in this circuit for at least the preceding three years. See Stoppelli v. United States, 183 F.2d 391, 396 (9th Cir. 1950). . Despite the apparent inconsistency, we have adopted the “reasonable hypothesis” test and, in the same opinion, stated that we apply no special rule in evaluating the sufficiency of circumstantial evidence. Elwert v. United States, 231 F.2d 928, 933, 935 (9th Cir. 1956). See also Hiram v. United States, 354 F.2d 4, 6 (9th Cir. 1965). Judge Hamley called attention to this implicit contradiction in Foster v. United States, 318 F.2d 684, 689-690 (9th Cir. 1963). . Most of our recent reversals for insufficiency of tlie evidence have occurred in narcotics cases involving the statutory presumption arising from possession. E. g., Davis v. United States, 382 F.2d 221 (9th Cir. 1967); Orozco-Vasquez v. United States, 344 F.2d 827 (9th Cir. 1965); Delgado v. United States, 327 F.2d 641 (9th Cir. 1964). The real difficulty in these cases is that the jury, usually because of inadequate instructions, has drawn inferences from the single fact of possession which are not sufficiently probable in light of all of the circumstances established by the proof. See Verdugo v. United States, 402 F.2d 599, 603-605 (9th Cir. 1968).