Nolan Reese appeals his conviction for distribution of cocaine base, in violation of 21 U.S.C. § 841 (a)(1). He contends that the district court erred when it denied his motion for discovery on the issue of selective prosecution. He also challenges the district court’s instructions to the jury on entrapment and raises other issues regarding evi-dentiary rulings and sentencing. We reverse and remand.

BACKGROUND

In 1993, Reese sold cocaine base to Tegi Rita Johnson and spoke to her about making further sales. Unbeknown to him, she was a confidential informant. Reese was ultimately indicted for distribution of cocaine base and was convicted on October 6, 1993.

After his conviction, but before his sentencing, Reese joined with twenty-three other defendants in a motion which sought dismissal of their indictments. They requested discovery and an evidentiary hearing relating to their claim that they had been selectively prosecuted on account of their race. As support, they submitted a study from the Federal Defender’s Office, declarations from geography professor Dr. John Weeks, a report published by the San Diego Association of Governments which detailed the racial composition of areas that law enforcement agencies commonly target, and statistics compiled by the California Department of Justice. The district court denied the motion on December 13, 1993, after a lengthy hearing. Reese was then sentenced, and this appeal followed.

JURISDICTION AND STANDARDS OF REVIEW

The district court had jurisdiction pursuant to 18 U.S.C. § 3231 . We have jurisdiction pursuant to 28 U.S.C. § 1291 .

A district court’s ruling on whether to order discovery relating to a claim of selective prosecution is reviewed for abuse of discretion. United States v. Armstrong, 48 F.3d 1508, 1512 (9th Cir.1995) (en banc).

“Whether a jury instruction properly states the law-of entrapment is a pure question of law subject to de novo review.” United States v. Lessard, 17 F.3d 303, 304 (9th Cir.1994).

DISCUSSION

A. Selective Prosecution.

In Armstrong , we held that the “proper standard for determining whether an adequate showing has been made by a defendant seeking discovery in connection with a selective prosecution charge” is the colorable basis test, not the prima facie showing test. 48 F.3d at 1510 . Armstrong adopted the colorable basis standard from United States v. Bourgeois, 964 F.2d 935, 939 (9th Cir.), cert. denied, — U.S. -, 113 S.Ct. 290 , 121 L.Ed.2d 215 (1992), with one substantial modification: it rejected the language in Bourgeois that a defendant’s evidence must meet a “high threshold” before a colorable basis is shown. 48 F.3d at 1513 .

The defendants in Armstrong had filed motions for discovery and for dismissal of the indictment based on selective prosecution. In support, the defendants submitted a study encompassing every case involving a charge under 21 U.S.C. §§ 841 and 846 that the Federal Public Defender’s office in the Central District of California had closed in 1991. In all twenty-four of those cases, the study showed, the defendant had been black. At the hearing, the government stated that it had no explanation for the disparity. Id. at 1511 . The district court granted the motion for discovery. The government filed a motion for reconsideration, relying upon evidence designed to show that its charging decisions were based upon race-neutral criteria. In response, the defendants submitted additional declarations from counsel and a newspaper article. The district court denied the motion for reconsideration. We affirmed the district court. We held that “discovery may be ordered when the evidence provides a colorable basis for believing that discriminatory prosecutorial selections have occurred. The existence of a colorable basis must be judged in light of all the evidence presented to the district court and not simply that offered by the defendant.” Id. at 1512 . We explained that the colorable basis standard “permits discovery when the defendants introduce some evidence tending to show the essential elements of selective prosecution and the government fails to explain it adequately.” Id. at 1514 . The essential elements of selective prosecution are that the prosecution “ ‘had a discriminatory effect and ... was motivated by a discriminatory purpose.’” Id. at 1513 (citation omitted).

The evidence submitted by the defendants here was similar to that presented in Armstrong and could support a determination that the colorable basis standard was satisfied. The evidence included a study, in the form of a declaration from Federal Defender’s Office investigator J. Patrick Loofbur-row, showing that of the 193 defendants charged with cocaine base offenses in the Southern District of California from 1989 to 1993, all but one were minorities, and most were black. Two declarations from Dr. John Weeks, a professor at San Diego State University, stated that the racial disparities in Loofburrow’s findings were statistically significant and could not reasonably be ascribed to chance. A report published by the San Diego Association of Governments suggested that law enforcement agencies concentrated cocaine base investigations in areas heavily composed of blacks, even though they had also identified at least one white area with high cocaine base activity. Finally, the defendants relied upon statistics compiled by the California Department of Justice, which tended to show that whites were more often prosecuted in state court than federal court for cocaine base offenses.

Under Armstrong , the Loofburrow declaration is essentially as strong as the evidence submitted in Armstrong and in some ways even stronger. See Armstrong, 48 F.3d at 1515 . As we said:

Id. Here, a similar study showed that all crack defendants but one represented by the Federal Defenders of San Diego over a five-year period were minorities, and the vast majority were black. That study alone could provide the necessary support for a colorable basis determination, with one possible exception. In Armstrong , the study may have directly correlated with the crimes with which the defendants were charged — i.e., cocaine base-dealing offenses under either 21 U.S.C. § 841 or § 846. However, nothing in Armstrong indicates that is so. Here, on the other hand, it appears that Loofburrow’s study could have encompassed cocaine base use and possession crimes, offenses with which none of the defendants were charged. The district court specifically relied upon that possibility when it discounted the relevancy of the Loofburrow study. Again, however, the form of the study is not different from Armstrong and there is no apparent reason for treating these statistics differently-

Under Armstrong , once a defendant has initially shown a colorable basis for the discovery he seeks, a district court must consider the rebuttal evidence offered by the government and determine whether, in light of it, a colorable basis for the claim still exists. Id. Here the government responded to the defendants’ showing by presenting its own compilation of statistics for the period, which it conceded it had not finished at the time of the hearing on the motion. That study preliminarily demonstrated a smaller disparity in the number of blacks prosecuted for cocaine base dealing versus the number of whites. The government also presented its own statistics concerning state prosecutions of cocaine base dealers. However, it does not appear that the statistical disparity is fully explained under either version of the relevant data. In short, we see no significant difference between the showing here and that in Armstrong .

Of course, we do not overlook the fact that in Armstrong the district court exercised its discretion to grant discovery and we found no abuse of that discretion. Here the district court exercised its discretion to deny discovery. Unfortunately, it did not have the benefit of our decision in Armstrong . Therefore, it relied upon the Bourgeois “high threshold” language when it ruled. As we have already pointed out, we have since disapproved of that approach. 48 F.3d at 1513 . The district court also relied upon the lack of direct evidence that similarly situated white persons were not being prosecuted in like numbers. Again, however, direct evidence is not required. Id. Finally, while the district court did not specifically rule on the discriminatory motive part of the test, it strongly indicated that statistics alone would not suffice to establish a colorable basis as to that part. In that the district court was also mistaken. Id. at 1513-14, 1519 .

It cannot be gainsaid that the court’s discretion must be exercised in accordance with Armstrong and that its failure to apply Armstrong’s standards requires reversal. See Griffith v. Kentucky, 479 U.S. 314, 328 , 107 S.Ct. 708, 716 , 93 L.Ed.2d 649 (1987) (“[A] new rule for the conduct of criminal prosecutions is to be applied retroactively to all eases ... pending on direct review or not yet final.”); United States v. Ortiz-Rivera, 1 F.3d 763, 768 (9th Cir.1993) (remanding for reconsideration in light of change in law between panel decision in a case relied upon by the parties and the court and en banc decision). We do not express an opinion regarding whether the district court should grant the defendant’s motion. Even though the record indicates that this case is much like Armstrong , that decision resides in the sound discretion of the district court. Armstrong, 48 F.3d at 1512 .

B. Entrapment.

Reese’s defense at trial centered on his claim that he was entrapped. He asserted that he had been importuned to sell drugs by the government informant, Johnson. He had no desire to do anything of the kind and told her so, he said. However, after she kept after him he finally obtained a few ounces of rock cocaine for her from his sister. His sister drove him to the rendezvous and handed him a bag which he handed to Johnson in return for money, none of which he personally kept. Others, who knew him well, testified that he was not a buyer or seller of rock cocaine.

Reese asked the district court to give a rather lengthy and convoluted instruction on entrapment. It refused and, instead, instructed the jury as follows:

The giving of that instruction was plain error. See United States v. Sterner, 23 F.3d 250, 252 (9th Cir.1994); United States v. Mkhsian, 5 F.3d 1306, 1310-11 (9th Cir.1993). Under Jacobson v. United States, 503 U.S. 540, 548 , 112 S.Ct. 1535, 1540 , 118 L.Ed.2d 174 (1992), the jury should have been instructed that it could convict the defendant only if the government proved beyond a reasonable doubt “that the defendant was disposed to commit the criminal act prior to first being approached by Government agents.” “Nor can we say that this error was harmless. Indeed, the essence of [Reese’s] defense was that he had been lured into criminal activity by [the] government [informant]. A properly instructed jury might have found that [Reese] was not predisposed to break the law prior to encountering [the informant].” Sterner, 23 F.3d at 252 . Thus, we are constrained to reverse Reese’s conviction.

CONCLUSION

Because the evidence here is so similar to that in Armstrong (indeed in some ways more compelling) and because the district court applied the wrong standards when it decided this case (before Armstrong was decided), we set aside the determination that the government need not provide discovery on Reese’s selective prosecution claim. In so doing we do not decide how the district court should exercise its discretion, nor do we preclude the taking of further evidence, nor, of course, do we speculate on the ultimate outcome of the proceedings.

We also reverse Reese’s conviction due to the error in instructing the jury on the entrapment defense.

REVERSED and REMANDED.

. The motion was heard by Chief Judge Keep. All other matters were heard by Judge Nielsen. . In so holding, Armstrong overruled United States v. Gutierrez, 990 F.2d 472, 476 (9th Cir.1993), to the extent that it required that a defendant seeking discovery on the basis of a selective prosecution claim offer some direct evidence showing that others similarly situated to him have not been prosecuted. 48 F.3d at 1513 n. 1, 1516. . The court also disapproved of the Weeks materials presented by the defendants on that ground. Table 4 of Weeks’ declaration does separate mere cocaine base possession for use crimes from sales-related crimes but his statistical analyses assume that dealers and users exist in the same percentages in the populations he considered. His reasons for that assumption are not entirely convincing. . According to the defendants’ figures, of 193 defendants prosecuted for cocaine base offenses in federal court from October 1989 to November 1993, approximately .5% (1) were white, 5.7% (11) were Hispanic, and 93.7% (181) were black. Using the government's numbers for 1989 to 1992 encompassing 56 defendants, about 1.7% (1) were white, 36% (20) were Hispanic and 62.5% (35) were black. It appears that the defendants may have counted certain defendants from Central America or the Caribbean as black, while the government considered them to be Hispanic. If those defendants are counted as Hispanic in the defendants’ study, their figures become approximately .5% (1) white, 12.4% (24) Hispanic, and 87% (168) black. .Those figures show that of 293 superior court dispositions for sale-of-cocaine-base offenses in 1992, 4% (14) were white, 62% (182) were Hispanic, and 31% (92) were black. . In light of reversal of the conviction on this ground, we do not consider Reese’s other claims of evidentiary and sentencing error, which may or may not arise at a new trial.