Opinion for the Court filed by Circuit Judge WILLIAMS.
On April 8, 1989 Raymond Powell approached an undercover police officer in the 3600 block of 6th Street in Southeast Washington and offered him “a 20 rock” (i.e., a $20 rock of cocaine base). The officer said that he wanted “a 50”, and Powell responded that he could “get the 50 from *132my man downstairs.” After giving a prearranged signal to his back-up team, the officer followed Powell down a flight of stairs into the basement of an apartment building. There he found three men awaiting him. One, Billy Williams, was holding a gun, though not pointing it at anyone in particular. The officer grabbed Powell and used him as a human shield until the backup team arrived. A sweep through the apartment revealed 13 rocks of cocaine base in concentrations averaging about 95%; an expert testified that the supply was consistent with street distribution rather than personal use.
Powell was convicted of possessing cocaine with intent to distribute in violation of 21 U.S.C. §§ 841(a), (b)(1)(C) and of using or carrying a firearm in the commission of a drug offense in violation of 18 U.S.C. § 924(c)(1),1 or (as to each) aiding and abetting the crime, see 18 U.S.C. § 2(a).2 For the firearms charge, the trial judge added five years to Powell’s sentence, as § 924(c) requires. While Powell’s attacks on the possession conviction are too weak to require discussion, we find the evidence insufficient to support the firearms charge and accordingly reverse that conviction.
Our standard of review is limited. We may reverse the jury’s verdict for insufficient evidence only if, allowing the government all reasonable inferences from the evidence, a reasonable mind could not conclude beyond a reasonable doubt that Powell was guilty. See United States v. Joseph, 892 F.2d 118, 125 (D.C.Cir.1989).
Our cases rule out liability as a principal. Though we have construed § 924(c) broadly to include both actual and constructive3 “use” of a firearm in the commission of a felony, see, e.g., United States v. Anderson, 881 F.2d 1128 (D.C.Cir.1989), the statute does have bounds. In United States v. Long, 905 F.2d 1572, 1578 (D.C. Cir.1990), the government made an argument similar to the one it urges today, which we paraphrased as follows: “[Defendant] was connected to the drugs; the distribution of the drugs was facilitated by the gun; since [defendant] thus derived benefit from the gun, he ‘used’ it.” Id. at 1576. Under such a view, anyone connected with drug distribution would violate § 924(c) whenever it turned out that an associate was using a firearm. This would be hard to square with Congress’s making firearm use a separate crime. Concerned that such strict liability “would obliterate any remaining limits on the meaning of the word ‘use,’ ” id., we noted that our cases upholding conviction under § 924(c)(1) as a principal were united by a “common theme”, namely, that possession (which we regarded as a necessary condition of “use”) was “indicated by one or more of several factors: close physical proximity to the firearm, possessory interest in the firearm, or dominion and control over the premises on which the firearm was located.” Id. at 1578.
The government has failed to establish any of those links here, or any that could reasonably be said to share a common theme with them. At oral argument it relied almost exclusively on Joseph, supra. In that case, however, the defendant’s sidekick (his younger brother) was carrying the gun in a bag as the two travelled together, so the jury could well have found that the defendant himself exercised dominion and control over the bag. 892 F.2d at 125-26. The government proved nothing comparable here. To the extent that the government invokes Joseph for the proposition *133that one is liable for another’s “use” of a gun when both are jointly involved in a drug operation, the theory merges into its aiding and abetting argument, to which we now turn.
It is common to state that liability as an accomplice encompasses acts of the principal that are a “ ‘natural and probable consequence’ of the criminal scheme the accomplice encouraged or aided.” W. La-Fave & A. Scott, Criminal Law § 6.8, at 590 (2d ed. 1986). Compare Pinkerton v. United States, 328 U.S. 640, 647-48, 66 S.Ct. 1180, 1184, 90 L.Ed. 1489 (1946) (in conspiracy cases, co-conspirators are liable for acts that, among other things, could be “reasonably foreseen as a necessary or natural consequence of the unlawful agreement”). The phrase “natural and probable consequences” by no means communicates just how likely the forbidden act must have appeared to the accomplice. It could signify any position within a broad range: for example, all acts with a substantial probability of occurrence (e.g., one chance in five); acts that are more probable than not to occur; acts of very high probability (e.g., 90%); and acts so likely that their occurrence is a practical certainty. Given the imperfection of human knowledge, the latter is the equivalent of knowledge; an accomplice “knows” an act will happen if he is “practically certain” it will. See Model Penal Code § 2.02, at 236-37 n. 13 (1985). Our review of the cases suggests that no court uses one of these degrees of probability for all contexts; each varies the requirement with the circumstances.
For provisions such as the current version of § 924(c), enhancing a sentence simply because a defendant had a firearm with him for possible use in an independent crime, the courts appear generally to have drawn the line at the upper end of the spectrum, insisting that the accomplice “knew” (or, perhaps, should have known)4 that the principal would carry a gun. Thus in United States v. Hamblin, 911 F.2d 551, 558 (11th Cir.1990), petition for cert. filed on other grounds sub nom. Jones v. United States, Feb. 26, 1991 (No. 90-7221), the court required actual knowledge, rejecting the idea that, as “it would be difficult to rob a bank without a weapon,” the jury could infer from that truism that the accomplice “had to know about the gun beforehand.” See also State v. White, 98 N.J. 122, 484 A.2d 691, 694-95 (1984) (accomplice guilty of robbery with firearms only if he knew or had reason to know that principal would possess or use the weapon); State v. Ivy, 119 Wis.2d 591, 350 N.W.2d 622, 629-30 (1984) (to be guilty of armed robbery as accomplice, defendant must have “known or believed” that principal would carry weapon); cf. Davies v. Director of Pub. Prosecutions, 1954 A.C. 378, 401 (where one crew of boys fights another and one boy pulls a knife and makes a fatal stab, others would be guilty of murder only if they intended “or at least contemplated an attack with a knife”). But see State v. Davis, 101 Wash.2d 654, 682 P.2d 883, 885-86 (1984) (law imposes “strict liability on all those involved in robbery which, by its very nature, generally requires use of weapons”).
We have found two cases involving the scope of knowledge required for accomplice liability under § 924(c) where the main crime was a drug violation; they split. In United States v. Morrow, 923 F.2d 427 (6th Cir.1991), the majority overturned the conviction for want of evidence that the accomplice encouraged the principal’s carrying of a gun, id. at 436, and even the dissent appeared to require that the accomplice clearly knew at the outset of the crime that his confederate was carrying a gun, id. at 443. In United States v. Johnson, 886 F.2d 1120, 1124 (9th Cir.1989), on the other hand, the court appeared to rest liability on the common understanding that the “drug industry ... is a dangerous, violent business,” id. at 1123, with guilt sustainable so long as the accomplice “could have reasonably foreseen" that the principal would carry a weapon. We see no reason why the test for ancillary gun liability should be *134laxer in drug trafficking cases than in bank robberies; possession of a gun is at most common in the former but is virtually essential in the latter.
A number of cases in our circuit — none of them involving the issue of ancillary gun possession — appear to employ a rather sweeping use of the “natural and probable consequences” tag. Thus, in United States v. Sampol, 636 F.2d 621 (D.C.Cir. 1980), we invoked it and then spoke vaguely of liability for “all acts ... which are in furtherance of the common design or plan to commit the felony”, id. at 676, quoting United States v. Heinlein, 490 F.2d 725, 735 (D.C.Cir.1973). But Sampol also relied explicitly on the doctrine of “transferred intent”, 636 F.2d at 676, a label for the rule that when the defendant seeks A’s death but kills B (as by bad aim), he is guilty of murdering B. The exact basis for the doctrine is obscure, see LaFave & Scott § 3.12(d), but appears to depend at least in part on the defendant’s moral evil being no less than if he hit the target. Id. Indeed, the doctrine seems necessary only because we conceive of murder as involving the intentional killing of a specific human being; a drug courier surely could not escape liability just because his principal suddenly switched plans and shipped Columbian instead of Peruvian cocaine. A Second Circuit decision, United States v. Blitz, 533 F.2d 1329, 1341-43 (2d Cir.1976); see also id. at 1346-47 (Mansfield, J., dissenting), seems to reflect a similar problem, with the panel splitting (in favor of liability) on whether participants in a large-scale market manipulation could be guilty of the specific sales by brokers unknown to them where they clearly should have known that some brokers would be making such sales.
United States v. Jones, 517 F.2d 176 (D.C.Cir.1975), is readily distinguishable as a case involving a variant of the doctrine of conditional intent. See LaFave & Scott § 3.5(d). There the court upheld defendant’s conviction for assault with a deadly weapon, committed in the course of an armed robbery. Though a confederate may have fired the shot, the defendant himself carried a gun, thus showing his readiness to resort to lethal violence under pressure. As he presumably intended use of a gun in the sort of circumstances that actually arose, he could not plausibly argue that he lacked the mental state needed for his conviction; indeed, his mental state was almost indistinguishable from the principal’s. United States v. Clayborne, 509 F.2d 473 (D.C.Cir.1974), is similar to Jones. The accomplice, knowing that the killer possessed a gun, held some valuables for him and thereby freed his hands both to attempt robbery on the victim and to inflict a fatal shot. Claybome, of course, besides being probably explicable on the same grounds as Jones, may also be an example of felony murder (as Jones itself would be if the victim had died).
In Sampol, Jones and Claybome each accomplice crossed a moral divide by setting out on a project involving either the certain or contingent use of deadly force. Once he had done so, the important probabilities involved not degrees of moral blameworthiness but mainly the resistance the venture might encounter (though, to be sure, there might also be variations in the principal’s ruthlessness). Though the accomplice would go free if by good fortune everyone were spared, it is hardly surprising that a court would hold him liable either for accidental variations in the identity of the victim or for a death or assault that follows readily when the feared contingency arises. But the courts’ references to “natural and probable consequences” in those contexts hardly mean that the phrase should entail the same degree of probability whenever an accomplice helps in the commission of one crime and the principal commits another. We find the standard of our fellow circuits more suitable, insisting that an accomplice at least know that a gun would be carried (or “used”, in the broad sense of § 924(c)) before he can be said to aid and abet that crime. This standard puts the accomplice on a level with the principal, requiring the same knowledge for both. See LaFave & Scott § 6.8, at 590-91; Model Penal Code § 2.06, Comment at 312 (1985); see also United States v. Greer, 467 F.2d 1064, 1068-69 (7th Cir. *1351972); United States v. Pope, 739 F.2d 289, 291-93 (7th Cir.1984).
We pause to note an intermediate class of cases, where the extra charge is not the carrying of a gun in the course of a bank robbery, but its active use — to put another’s life in jeopardy. See 18 U.S.C. § 2113(d). If we correctly discerned the relation between the gun-carrying cases and the Sampol/Jones group, we would expect the courts to insist on knowledge that someone would bring a gun, but not that the holder would use it actively — a matter likely to depend on how the scene unfolded. In fact, the cases present a mixed picture. In United States v. San-born, 563 F.2d 488, 491 (1st Cir.1977), for example, the court said that “the Government must show that the accomplice knew a dangerous weapon would be used or at least that he was on notice of the likelihood of its use.” (Emphasis added.) The court found the evidence ample. The accomplice himself carried a gun as he manned the getaway car, and he had been with the principal for a day-long drive to the site just before the robbery; one “so closely associated with the venture could not fail to know what would be the central question in any robbery: how the robbers were to force the bank's employees to part with the money.” Id. at 490. Though the court found error in the charge, it said it would have been enough to tell the jury it must find “notice of the likelihood that a gun or other dangerous weapon would be used.” Id. at 491. The tension in the opinion between knowledge and mere notice of likelihood seems an understandable reaction to the mixture of the two elements — the carrying of the gun (there, a practical certainty) and the exact mode of its use (a matter of speculation). United States v. Short, 493 F.2d 1170, 1172 (9th Cir.1974), required proof that the accomplice “knew that [the principal] was armed and intended to use the weapon,” as did United States v. Jones, 592 F.2d 1038, 1041-42 (9th Cir.1979), but because of the ambiguity in the word “use” (to explicitly put someone in jeopardy, or merely to be ready to do so if necessary?), the cases may not require knowledge that the principal would actually brandish the gun. In any event, the cases are quite consistent with our requirement that for § 924(c) liability the accomplice must have known to a practical certainty that the principal would be carrying a gun (using “carry” as a stand-in for all variants of guilt under § 924(c)).
The record here fails to show that knowledge. The government points to the testimony of its expert witness, who said that he (personally) had “frequently” “come across” the use of guns in connection with drug distribution. Tr. 128. This, coupled with his explanation of a drug dealer’s purpose in wielding a gun, was useful to erase any doubts jurors may have had that Williams possessed and flourished the gun in order to advance his drug trafficking. (It also helped prove that the drugs were possessed for distribution purposes.) But it falls considerably short of showing that someone in Powell’s position would know that Williams would be carrying a gun.
The government invokes several cases, United States v. Bonner, 874 F.2d 822, 824 (D.C.Cir.1989); United States v. Payne, 805 F.2d 1062, 1065 (D.C.Cir.1986); United States v. McDowell, 762 F.2d 1072, 1075 (D.C.Cir.1985), in which we acknowledged a common-sense link between guns and drugs. In the first case we cited the link as evidence that police officers did not act unreasonably when, after waiting for suspected drug dealers to answer their knocks on the door, they rammed it down instead of waiting vulnerably outside. Obviously the dangers from a firearm are grave enough that even a modest chance of a gun’s being present in the dwelling (e.g., 10%) may justify such conduct. In the latter two cases, we upheld the admission of evidence of a defendant’s possession of (respectively) a firearm and a bulletproof vest to prove intent to distribute the drugs defendant was shown to possess. None of the three decisions depended on an empirical showing that the drug trade so invariably entails gun use that participants in the one can be said to know that they are involved in the other. In fact, we once *136endorsed a study — now dated but still suggestive — that “showed that 10% of drug suspects arrested in Washington included in the survey were armed.” United States v. White, 648 F.2d 29, 35 n. 29 (D.C.Cir. 1981). While the frequency is surely greater among higher ranking retailers (such as Williams), we cannot say by how much.
Moreover, even if guns were shown to be a part of an overwhelming majority of drug operations, not all drug traffickers — or drug trafficking contexts — are alike. We suspect, for example, that many a college dealer gets along without firearms. Without evidence of the prevalence of guns in a particular context, no jury can reasonably infer that someone operating in that context knows that his associates will carry a gun.
The government suggests that Powell and Williams had established such a close criminal relationship that, given Williams’s use of the gun at the arrest, a jury could have inferred that Powell must have known of its use in the time period charged. Cf. Joseph, 892 F.2d 118 (gun-holder was brother, subordinate, and travelling companion of defendant charged with § 924 offense); United States v. Douglass, 780 F.2d 1472, 1476 (9th Cir.1986) (discussing importance of defendants’ “long association with” co-conspirator who possessed gun). Yet as evidence of this kind of relationship, the government points only to Powell’s having told the undercover officer that he could get the drugs from “my man downstairs”. See Tr. 47. But the term “my man" is quite consistent with even the early stages of a dealer-runner relationship.
In sum, the government has failed to offer specific evidence that this defendant knew to a practical certainty that his drug-peddling associates would be carrying a gun in connection with their joint venture.
The parties have addressed only the sentence enhancement that Congress explicitly provided in § 924(c). The sentencing guidelines supply an alternative route to enhancement, directing the judge to add two levels toward the sentence for a drug charge when the government shows that the use of firearms was “reasonably foreseeable”. See Sentencing Guidelines §§ 2D1.1(b), 1B1.3(a) and comment; and see, e.g., United States v. Burke, 888 F.2d 862, 868 (D.C.Cir.1989); United States v. Barragan, 915 F.2d 1174, 1178-79 (8th Cir. 1990); United States v. Garcia, 909 F.2d 1346, 1349-50 & n. 1 (9th Cir.1990). This weaker standard reflects the general understanding that less proof is needed for an augmented sentence within a legislatively determined range than for an enhancement expanding the range. See Burke, 888 F.2d at 868. As the issue has not been raised, we need not address the exact meaning of the standard.
We affirm Powell’s conviction for possession with intent to distribute and reverse his conviction for using or carrying a firearm in the drug venture.
So ordered.