concurring, with MOORE, J., joining in Part I only.
I.
We concur with Judge Gilman’s opinion, and agree with its reasoning on most aspects of this complex case. Nonetheless, we write separately because we are persuaded by Murray’s and Washpun’s argument that the officer testimony that trained canines reacted positively to currency found on them should have been ruled inadmissible. Both defendants contend that this “dog-sniff’ evidence is inherently unreliable because it does not necessarily indicate drug activity on their part, citing studies finding that anywhere from seventy to ninety-six percent of United States currency is tainted with narcotics.1 They presented these statistics to the district court through their motions in limine and oral arguments, as well as through voir dire of government witnesses.
We agree that this dog-sniff evidence was inherently unreliable and that the court abused its discretion in admitting it. In recent years, this court and others have expressed skepticism regarding the probative value of evidence that a dog detected narcotics traces on currency.2 In United States v. $5,000 in United States Currency, 40 F.3d 846 (6th Cir.1994), this court held that the evidentiary value of the narcotics dog’s alert was minimal, and “insufficiently indicative of probable cause.” Id. at 848-49. The court cited cases and studies indicating that up to ninety percent or more of bills test positive for traces of cocaine. See id. at 849. This conclusion followed a previous panel which had found that similar dog-sniff evidence had only weak probative value. See United States v. $53,082.00 in United States Currency, 985 F.2d 245, 250 n. 5 (6th Cir.1993). Other circuits have similarly doubted the utility of such evidence. See, e.g., United States Currency, $30,060.00, 39 F.3d at 1043 (concluding that statistics showing widespread currency contamination greatly diminishes the probative value of positive dog sniffs of money, and that continued reliance of courts and law enforcement officers on such evidence is “logically indefensible”) (citation omitted); United States v. $191,910.00 in United States Currency, 16 F.3d 1051, 1062 n. 21 (9th Cir.1994) (noting that “[i]n recent years, courts have increasingly questioned the reliability of dog alerts” on currency); Jones v. Drug Enforcement Administration, 819 F.Supp. 698, 719, 720 (M.D.Tenn.1993) (concluding that because contaminated currency is widespread, evidence of a “narcotic-trained dog’s ‘alert’ to the currency is of extremely little probative weight”).
Courts generally have not translated these doubts in the forfeiture realm into an outright prohibition under Fed.R.Evid. 403. See, e.g., United States v. Saccoccia, 58 F.3d 754, 778 (1st Cir.1995) (affirming admission of dog-sniff evidence); United States v. Akins, 995 F.Supp. 797, 814 (M.D.Tenn.1998) (allowing dog-sniff evidence and concluding that it was not “unduly prejudicial”). Nevertheless, such cases also involved unique circumstances. *316In Saccoccia, the appellant had not presented findings to the district court regarding the non-reliability of such evidence, so those materials could not inform the district court’s decision. See 58 F.3d at 777 n. 19; cf. United States v. Carr, 25 F.3d 1194, 1202 n. 3 (3d Cir.1994) (declining to take judicial notice that nearly all currency contains detectable traces of narcotics). And in Akins, illegal drugs were found in the bag of the defendant whose money was also sniffed by a narcotics dog. See 995 F.Supp. at 814. Indeed, when there were other factors diminishing the reliability of a separate dog sniff, the Akins Court concluded that the evidence was unduly prejudicial and did not allow its admission. See 995 F.Supp. at 814 (finding limited probative value and unfair prejudice for a sniff of money which was contained in a drawer in a DEA interdiction room).
In Carr, Judge Becker voiced perhaps the strongest case for finding dog-sniff evidence inadmissible in particular cases:
If any of the many studies [regarding currency contamination] is valid, then the fact that a dog alerted to a large number of bills in United States currency which has circulated in a major metropolitan center (at which the studies are directed) is meaningless and likely quite unfairly prejudicial, see Fed. R.Evid. 403, and evidence thereof should have been excluded. Although having been directed to many of the studies ..., the government in its brief has not disputed the validity of any of the studies mentioned above [nor] pointed to any countervailing studies.... It is thus my considered opinion that the fact that numerous studies by governmental and private agencies, studies which stand un-refuted, strongly suggest that a trained canine will alert to all bundles of used currency does not permit the jury to draw a reasonable inference that the person in prior possession of such currency was a drug trafficker or associated with one. Indeed, I am inclined to the view that the information now available establishes a strong presumption against the admissibility of evidence of a canine’s alert to currency, and that the government can rebut that presumption only if it first clearly and convincingly establishes, outside the presence of the jury, the relevance and non-prejudicial character of the offered evidence.
25 F.3d at 1216-17 (Becker, J., concurring in part, dissenting in part) (footnotes omitted) (emphasis supplied in original).
Given the unrebutted statistical studies in this and other cases, we find Judge Becker’s view compelling. We believe that courts should generally presume against the admissibility of evidence that a dog detected narcotics on currency unless the government offers other evidence showing a direct nexus between illegal narcotics, the currency in question, and the defendant. Further, when circumstances of the dog-sniff detection in any way cast doubt on the reliability of that evidence, such as in Akins, we believe courts should find such evidence inadmissible. Under such a presumption, we believe that the facts of both Murray’s and Washpun’s arrests militated for exclusion of the dog-sniff evidence in question. First, as in Carr, the government did not attempt to rebut the contamination studies. Indeed, one officer testified that she was aware of studies showing that as much as seventy to ninety percent of currency is contaminated with some ' amount of controlled substances, J.A. at 1034-35, and none of the witnesses who testified as to the dogs’ training countered those findings. Further, in neither case was there a nexus between the currency found and illegal narcotics. The dog sniff in Murray’s case followed a traffic stop, and although Murray was found with a large amount of cash, he possessed no narcotics. Similarly, the dog sniff in Washpun’s case came after a routine traffic stop of a ear in which Wash-pun was a passenger. A dog “reacted” to currency in the glove compartment, as well as to money which had only seconds before *317been removed from Washpun’s person. Nevertheless, officers found no drugs in Washpun’s car. Given the studies cited above, we find it disquieting that prosecutors utilized this dog-sniff evidence alone to create the inference that these individuals were engaged in illegal narcotics activity at those times. Nor are we comforted, as the district court is, by the fact that the jurors were aware of and could “discount” the possible problems with dog-sniff evidence. Federal Rule of Evidence 403 explicitly recognizes that certain evidence will “confusfe] the issues [] or mislead[] the jury,” and it is the judge’s duty to exclude such evidence when the potentiality of those effects substantially outweighs the evidence’s probative value. Fed. R.Evid. 403. We believe that in this case, the dog-sniff evidence carried just this risk.
The Government argues that defendants’ citation of statistical studies was insufficient to justify our conclusion that the evidence in question is inherently unreliable. We disagree. In addition to defendants’ citation to these studies in motions in limine, oral argument, and in examining witnesses, as well as this Court’s past citation of these studies, the Government’s own witness acknowledged the data from the studies, and the district court recognized that “the studies that show that the residue is on ninety percent of the currency [raise] an interesting issue.” J.A. at 1049. Despite the fact that defendants had plainly raised the studies in challenging the dog-sniff evidence, the Government made no effort to rebut those studies at trial, and again failed to do so on appeal. Indeed, the Government acknowledged the problem identified by the studies when it echoed the very presumption against such dog-sniff evidence that we have couched here; specifically, the Government conceded to the district court that if “this were the only evidence we had against Mr. Murray in this case, I would agree with [counsel] that it was more prejudicial than probative.” J.A. at 1046. Because there was no other evidence of drug activity at the particular stops in question, we believe applying the presumption is appropriate.
The Government further argues that our conclusion is in error because its witnesses discussed their dogs’ reliability in detecting cocaine, and explained that on several occasions, the dogs in question did not react to “general population” money or money withdrawn from a bank. But this testimony as to individual dogs’ ability does nothing to rebut the underlying problem highlighted by the studies — that the widespread contamination of money means that a dog’s detection of cocaine on currency provides little support for the inference that the possessor of that money was engaged in illegal narcotics activity. Stated simply, the problem raised by the studies is the reliability of the contaminated money as evidence of wrongdoing, not the ability of the drug-sniffing dog. The witnesses’ testimony about their dogs’ general training and reliability, including that on several occasions the dogs did not react to “general circulation” money, did not assuage this concern. Indeed, aspects of their testimony — such as the fact that police simply re-place confiscated drug money (on which dogs have previously detected narcotics) into general circulation without “cleaning” it, J.A. at 1180-81, and that one contaminated bill in a stack of bills can trigger a dog’s response, J.A. at 1179— only bolstered the defendants’ argument. The district court clearly sensed that the officer-witnesses’ testimony had not responded to the defendants’ underlying concern: “clearly the cross-examination of [Parsons] on voir dire about success rates was kind of — communicative-wise they weren’t communicating. Officer Parsons doesn’t think that way; Mr. Doele and perhaps maybe lawyers think that way on this particular issue.” J.A. at 1048. Despite this conclusion and its acknowledgment that the studies were “interesting,” the court concluded that it did not find the evidence “highly prejudicial based upon the fact that I think a preliminary showing *318has been made that this dog has a fair degree of reliability in what the dog does.” J.A. at 1050 (emphasis added). But once again, the fact that the dog and dog-trainer reliably detect narcotics is not relevant. The unrebutted studies show that the detection of narcotics on currency is an unreliable indicator of whether a particular defendant is involved with illegal narcotics, even if the dogs’ ability to detect narcotics (doing “what the dog does”) is perfectly reliable.
Despite our concern on this issue, we do not think admitting the dog-sniff evidence was reversible error. Even without that evidence, there is substantial other evidence linking both Murray and Washpun to the conspiracy and to specific criminal actions.
II.
Second, I am uneasy with the government’s use of the challenged group photographs in this case. To an undiscerning eye, the use of the photographs showing the defendants in a relaxed social setting may seem to be of no evidentiary consequence. To those who have been victims of the subtleties of race, however, the conditioning effect of such a display is most apparent. In the context of this case, with the racial implications resulting from an all-white jury and an all-black set of defendants, I worry that the photographs, introduced as early in the trial as they were, likely had an improper, not-so-benign racial conditioning effect. Evidence relating to illicit relationships between defendants which might otherwise be viewed with skepticism may subconsciously have been granted a degree of credibility by virtue of the photographs at issue. In other words, they introduce more prejudice than probative value. Once again, however, given the broader evidence linking defendants together in this conspiracy, I do not believe the use of these photographs had an impact on the outcome of this trial. Then-introduction was thus harmless error.
. Defendants cite two reasons that such a large percentage of currency is tainted with narcotics. First, when currency is run through a mechanized counter at a bank, narcotics contained on some of the currency gets into the counter and is transferred to other currency. Second, the ink on currency bonds with the narcotics. This argument echoes other courts’ and studies' conclusions regarding "contaminated” money. See, e.g., United States v. United States Currency, $30,060.00, 39 F.3d 1039, 1042 (9th Cir.1994).
. The issue of the reliability of dog-sniff evidence emerges in two different contexts: 1) as here, whether it should be allowed as evidence, and 2) whether it is sufficiently indicative of probable cause for forfeiture purposes.