Esperanza Aguilar-Aranceta ("Aguilar-Aranceta") was convicted for possession of cocaine with intent to distribute, in violation of 21 U.S.C. Sec. 841(a)(1). She now appeals. For the following reasons, we reverse.

On September 22, 1990, Aguilar-Aranceta went to pick up two parcels at the United States Post Office in Old San Juan, Puerto Rico. Two yellow slips had been left in her mail box prior to that date notifying her that there were two registered parcels addressed to her at the post office.1 Addressing Aguilar-Aranceta in English, the window clerk at the post office counter asked for identification to verify the signature on the two slips. Aguilar-Aranceta responded by immediately retrieving a Puerto Rican driver's license from her purse. After verifying the signatures, the window clerk brought her two packages with return addresses from Medellin, Colombia. Both packages were addressed to Esperanza Aguilar, P.O. Box 5739. Box 5739 was rented to Aguilar-Aranceta. Aguilar-Aranceta pointed at the return addresses on the packages and stated "no me (sic) family," to which the attendant replied that it was up to her if she wanted to take them or not. The window clerk left the packages on the counter and once again the defendant said "no me (sic) family," and once again the clerk replied that it was up to her if she wanted to take them or not. Defendant then proceeded to take the two packages. Immediately upon exiting the lobby of the Post Office defendant was detained and placed under arrest. The two packages she was carrying had been intercepted by a mail specialist earlier and found to contain approximately 224 grams of cocaine. Aguilar-Aranceta never opened the packages.

On September 26, 1990, a federal grand jury in San Juan, Puerto Rico, returned a two count indictment against defendant, charging her with unlawful possession of approximately 224 grams of cocaine with the intent to distribute, in violation of 21 U.S.C. Sec. 841(a)(1) (count one); and with importation of the same cocaine to the United States from Medellin, Colombia, in violation of 21 U.S.C. Sec. 952(a) (count two). Aguilar-Aranceta entered pleas of not guilty as to both counts. The jury returned a verdict of not guilty as to the importation count, but was unable to reach a unanimous verdict as to the count alleging possession with intent to distribute. The district court declared a mistrial as to count one and subsequently granted the government's request for a new trial.2

After a second jury trial in June 1993, Aguilar-Aranceta was convicted for possession of cocaine with the intent to distribute. Aguilar-Aranceta now appeals.

To convict Aguilar-Aranceta under 21 U.S.C. Sec. 841(a)(1), the government was required to show beyond a reasonable doubt that she knowingly possessed a controlled substance with the intent to distribute.3 United States v. Bergodere, 40 F.3d 512, 518 (1st Cir.1994), cert. denied, --- U.S. ----, 115 S. Ct. 1439, 131 L. Ed. 2d 318 (1995). The government presented evidence concerning Aguilar-Aranceta's prior conviction for possession of cocaine. The district court admitted this evidence as relevant to the issue of whether Aguilar-Aranceta was in knowing possession of cocaine when she was arrested at the Old San Juan Post Office in 1990. Aguilar-Aranceta contends that all evidence pertaining to her prior conviction serves no other purpose than to demonstrate a propensity for criminal activity and should therefore have been excluded under Federal Rule of Evidence 404(b).

This circuit is no stranger to the problems surrounding the admissibility of extrinsic act evidence under Rule 404(b).4 See, e.g., United States v. Guyon, 27 F.3d 723, 728-29 (1st Cir.1994); United States v. Fields, 871 F.2d 188, 195-99 (1st Cir.1989); United States v. Mateos Sanchez, 864 F.2d 232, 234-38 (1st Cir.1988); United States v. Oppon, 863 F.2d 141, 144-48 (1st Cir.1988). We have adopted a two-part test to determine the admissibility of such evidence. Oppon, 863 F.2d at 146. First, the trial judge must determine whether the evidence in question is offered for any purpose other than solely to prove that the defendant had a propensity to commit the crime in question. United States v. Garcia, 983 F.2d 1160, 1172 (1st Cir.1992); United States v. Moccia, 681 F.2d 61, 63 (1st Cir.1982). That is, the judge must determine whether the evidence has some "special" probative value. United States v. Arias-Montoya, 967 F.2d 708, 709 (1st Cir.1992). Prior bad acts may be "specially relevant" if they are probative of motive, opportunity, intent, preparation, plan, knowledge, identity, or absence of mistake or accident. See, e.g., Guyon, 27 F.3d at 728; Garcia, 983 F.2d at 1172.

If the judge is satisfied that the proffered evidence has "special relevance," the focus shifts to the second part of the test, which applies Rule 403 to determine whether the probative value of the evidence is "substantially outweighed by the danger of unfair prejudice."5 Fed.R.Evid. 403; Guyon, 27 F.3d at 728-29; Garcia, 983 F.2d at 1172. On appeal, we review the Rule 404(b) determination for abuse of discretion. Guyon, 27 F.3d at 728-29; United States v. Cassiere, 4 F.3d 1006, 1022 (1st Cir.1993); see also Garcia, 983 F.2d at 1172.

Applying these principles to the facts at hand, we first must determine whether the district court abused its discretion in finding that Aguilar-Aranceta's 1986 conviction for misdemeanor possession of cocaine was specially relevant to the issue of knowledge in the present case. Although the law regarding prior bad acts as evidence of knowledge is well settled, it is complex and merits some discussion.

In prosecutions for "possession" offenses,6 the central issue is often whether the defendant was in knowing possession. The knowledge element is difficult to prove, and defendants commonly claim that they were merely innocent bystanders or unwitting participants. Where the evidence is susceptible to the explanation that the acts alleged to constitute the crime were innocently performed and the crucial issues of intent and knowledge are keenly disputed, we have held that it is within the judge's discretion to permit the government to introduce evidence of prior similar offenses to demonstrate the unlikeliness that the defendant was merely an innocent and unknowing bystander. For example, in United States v. Spinosa, 982 F.2d 620 (1st Cir.1992), we held that evidence concerning the defendant's history of cocaine possession and dealing was probative of his knowledge and intent to participate in cocaine transaction charged because it directly contradicted his defense that he was involved only by accident or mistake.

The justification often advanced for admitting evidence of other crimes to prove knowledge is that no inference as to the defendant's character is required. See United States v. Ferrer-Cruz, 899 F.2d 135, 138 (1st Cir.1990); see also 22 Charles A. Wright & Kenneth A. Graham, Jr., Federal Practice and Procedure, Sec. 5245 (1978). Rather, the probative value emanates from the law of probabilities. We consider the following example illustrative of the permissible inferential chain:

The fact that an illegal alien was previously found stowed away under the hood of defendant's car does not prove that a second alien found there did not creep in undetected while the car was parked. However, it does seem unlikely that the same person could be twice victimized in this fashion, particularly when a reasonable person who had previously gotten into trouble in this way would probably take care to see that he had only an engine under the hood before heading for the border.

22 Wright & Graham, supra, Sec. 5245. Along these lines, we also have explained that jurors might permissibly hypothecate that the defendant's repetitive involvement in criminal conduct is unlikely to have left her oblivious to the true character of the acts in question. See, e.g., Ferrer-Cruz, 899 F.2d at 138 (noting that "[s]ince one who has previous experience with drugs is more likely to see 'car switching' as part of drug sale technique than one who has no such experience, ... the inferences at issue do not involve character"); United States v. Simon, 842 F.2d 552 (1st Cir.1988); United States v. Estabrook, 774 F.2d 284, 288 (8th Cir.1985).

"There is, however, a danger that the supposed inferences to knowledge will be obscured by the forbidden inference to propensity, particularly in cases in which the theory of knowledge is the probability that the defendant would have obtained knowledge in the course of repetitive involvement in criminal conduct." Wright & Graham, supra, Sec. 5245. It is, therefore, important to explain what inferences are impermissible. It is impermissible to suggest that a defendant's prior conviction for possession of cocaine somehow makes her more likely to have acted in a similar manner in the context of the present possession charges. Similarly, a prosecutor may not argue that the defendant's prior conviction for a drug offense demonstrates her propensity for involvement in drug trafficking.

Moving to the present case, we note that the circumstances surrounding Aguilar-Aranceta's 1986 conviction are similar to those of the present charges in several significant aspects. We have held that similarity between the prior act and the current charges is often the predominant factor or touchstone in the test for special relevance as to defendant's knowledge. See United States v. Arias-Montoya, 967 F.2d at 712-13 (summarizing circumstances which would constitute "special relevance," including that prior bad acts (1) closely mirrored the newly charged crime, (2) were part of a common scheme or a course of continuous dealing, or (3) provided the context for the newly charged crime); see also United States v. Hadfield, 918 F.2d 987, 994 (1st Cir.1990) (upholding admission of evidence concerning prior drug involvement in a drug trafficking case because the past actions were "proximate in time and closely allied with the type of crimes for which appellants were being tried"), cert. denied, 500 U.S. 936, 111 S. Ct. 2062, 114 L. Ed. 2d 466 (1991). Here, both cases involved packages containing cocaine that had been shipped to Aguilar-Aranceta's post office box at the Old San Juan Post Office. In both instances, the packages were addressed to Aguilar-Aranceta with return addresses indicating that they had been sent from Medellin, lombia. The cases differ with regard to the circumstances of arrest. In 1986, the authorities followed Aguilar-Aranceta from the Old San Juan Post Office and arrested her at her home. In the present case, Aguilar-Aranceta was arrested before she left the Post Office Building. In both cases Aguilar-Aranceta had not opened the packages at the time of her arrest. In the 1986 case, the unopened packages had been in her possession for several hours.

With regard to whether the prior conviction clears the special relevance hurdle, it is a close call. We are particularly concerned with the four-year period between her prior conviction and the facts leading to the present charges as well as the fact that the packages were unopened in both instances. Nevertheless, we do not find that the district court abused its discretion in concluding that the 1986 conviction is so similar that it is relevant to the issue of knowledge in the present case. A jury could have permissibly made the following inferential analysis: many people in the general population have little or no knowledge of how narcotics traffickers use the mail system to ply their trade, and for this reason might unwittingly accept two packages sent to them from an unfamiliar address in Medellin, Colombia. Aguilar-Aranceta's prior conviction for possession of cocaine that had been mailed to her from Medellin, Colombia suggests that she cannot plausibly make this claim. See United States v. Nickens, 955 F.2d 112, 124-25 (1st Cir.) (holding that where defendant claimed to have been an innocent dupe with regard to cocaine found in his luggage, his prior narcotics conviction was relevant to issue of knowledge because a jury might permissibly infer that someone who has experience selling cocaine is more likely to know how drug smugglers operate), cert. denied, --- U.S. ----, 113 S. Ct. 108, 121 L. Ed. 2d 66 (1992). Of course, her prior conviction does not conclusively prove that Aguilar-Aranceta knew that the packages contained contraband. Nevertheless, we think that the jury permissibly could have inferred that someone with a previous mail-related narcotics conviction would be reluctant to again accept mysterious packages from Colombia, and that Aguilar-Aranceta's failure to reject the packages given her prior experiences bears on the crucial issue of knowledge. We think that these inferences might be even more plausible given that Aguilar-Aranceta claims to have been an unwitting pawn in the events that led to her guilty plea in 1986.7 Accordingly, we find that the district court did not abuse its discretion in finding that Aguilar-Aranceta's prior conviction was specially relevant to the issue of knowledge.

We now consider whether the district court should nevertheless have excluded the evidence under Rule 403. The tenets of Rule 403 balancing are familiar and often quoted: "If the evidence brings unwanted baggage, say, unfair prejudice or a cognizable risk of confusing the jury, and if the baggage's weight substantially overbalances any probative value, then the evidence must be excluded." United States v. Rodriguez-Estrada, 877 F.2d 153, 155 (1st Cir.1989). We note, however, that, "[b]y design, all evidence is meant to be prejudicial; it is only unfair prejudice which must be avoided." Id. at 156. Moreover, "[t]he phrasing of Rule 403 makes it clear that the discretion to exclude does not arise where the balance between the probative worth and the countervailing factors is debatable; there must be a significant tipping of the scales against the evidentiary worth of the proffered evidence." Wright & Graham, supra, Sec. 5221 at 309-10. Accordingly, we review only for abuse of discretion. United States v. Desmarais, 938 F.2d 347, 351 (1st Cir.1991).

Aguilar-Aranceta argues that the legitimate probative value of her prior conviction, if any, is completely overshadowed by the danger of unfair prejudice. We agree. The prior conviction is probative on the issue of knowledge only in an attenuated manner, dependent on the following "once burned, twice shy" chain of inferences: (1) someone who has a previous conviction stemming from the receipt of cocaine-laden packages would likely be reluctant to innocently and unwittingly accept mysterious packages from Colombia; and (2) the fact that Aguilar-Aranceta did in fact accept the packages despite her prior conviction suggests that she knew their contents. While in some circumstances this inferential chain might be strongly probative, the circumstances here render it of limited value. First, we note the remoteness in time of her prior conviction. Common sense dictates that the time span between the events bears directly on the probative weight of the prior conviction vis-a-vis the government's "once burned, twice shy" argument. Cf. United States v. Lynn, 856 F.2d 430 (1st Cir.1988) (noting that six year period between the prior conviction and the instant offense significantly diminishes the probative value of the prior conviction). Second, we note that Aguilar-Aranceta apparently spoke little or no English. She testified that she was expecting a letter from her sister and became confused when the window clerk produced packages with unfamiliar return addresses. She testified further that she eventually accepted the packages because the window clerk kept insisting that the packages were hers even though she repeatedly tried to explain that the packages were not from her family. Third, we note that the window clerk was aware that Aguilar-Aranceta would be arrested if she took the packages, which suggests that he might have been especially zealous in encouraging her to take the packages. We think these circumstances limit the probative value of the prior conviction evidence.

Against the marginal relevance of the prior conviction evidence, we weigh the danger that it unfairly prejudiced the jury. Here, we cannot escape the conclusion that the most powerful inference that the jury was likely to make from the prior conviction is also the forbidden one: that because she was previously convicted under nearly identical circumstances, she must be guilty here. The specter of impermissible character evidence is likely to have significantly overshadowed any legitimate probative value. We find this especially likely given that there was virtually no other evidence suggesting that Aguilar-Aranceta was in knowing possession of cocaine. Leaving aside her previous conviction, the evidence pertaining to the defendant's state of mind was equivocal. The government points out that although Aguilar-Aranceta was poor and was able to receive mail at her home, she kept a post office box at the Old San Juan Post Office. Nevertheless, Aguilar-Aranceta was apparently in no hurry to get the packages. In fact, nine days passed between the time the first set of claim slips were placed in Aguilar-Aranceta's box and the time that Aguilar-Aranceta went to pick up the packages. Similarly, her behavior at the counter was of uncertain significance. The window clerk could only state that since she did not expressly refuse the packages, he continued to leave them on the counter. The government's argument that Aguilar-Aranceta's demonstrated reluctance was merely a smokescreen to conceal her knowledge of the contents of the packages is dubious, especially in light of the fact that the window clerk could have taken back the packages at any time. Perhaps because the government's evidence was weak with respect to the current charges (particularly so because it arrested Aguilar-Aranceta before she had a chance to open the packages), it focused the jury's attention on her prior conviction. In any case, we think that the amount of evidence introduced with respect to her prior conviction dramatically increased the likelihood that the jury convicted Aguilar-Aranceta on the basis of the 1986 conviction alone rather than on a fair assessment of the evidence.

The gravity of the unfair prejudice is even more apparent when one considers that the prior conviction was not merely one part of the government's attempt to establish Aguilar-Aranceta's state of mind. It was the entire case. The offense of possession with intent to distribute cannot be "established by proof merely that a package containing drugs was mailed from outside this country and was received and opened by the addressee of the package inside this country. The threat this would pose to innocent victims of mere mistake or actual set-ups is obvious." United States v. Samad, 754 F.2d 1091, 1096 (4th Cir.1984). This logic applies with equal or greater force to someone with a previous conviction for narcotics trafficking. In essence, admitting Aguilar-Aranceta's prior conviction allowed the jury to convict her upon facts that would likely have been insufficient to convict a similarly situated defendant without a prior conviction. In sum, we think that evidence of Aguilar-Aranceta's prior conviction should have been excluded because its marginal probative value, coupled with the scarcity and equivocal nature of the other evidence relating to the Aguilar-Aranceta's state of mind, created an unacceptable risk that the jury would assume that Aguilar-Aranceta had a propensity for narcotics trafficking and convict on that basis alone.

A non-constitutional evidentiary error under Rule 404(b) will be treated as harmless only if it is "highly probable" that the error did not contribute to the verdict. Arias-Montoya, 967 F.2d at 714 (citing United States v. Garcia-Rosa, 876 F.2d 209, 222 (1st Cir.1989)). Given the scarcity of other evidence pertaining to Aguilar-Aranceta's state of mind, we think it is highly probable that the evidentiary error did affect the verdict. Accordingly, we find that the district court's error was not harmless.8

Reversed.