OPINION
GOODWIN, Circuit Judge.
Two city police officers appeal a judgment for damages following a jury verdict in a 42 U.S.C. § 1983 action brought against the city and the individual officers. We affirm.
I. Introduction and Procedural History
Jeffrey Allen Grant was arrested and charged with a series of highly publicized rapes and related felonies (the “Belmont Shores rapes”) that occurred over eighteen months in the City of Long Beach. When forensic evidence found at several crime scenes failed to match Grant’s DNA, the prosecutor dropped all charges and released Grant from jail, where he had been sitting for over three months awaiting trial. Grant then sued the City of Long Beach, the Long Beach Police Department, and the two police officers that spearheaded the investigation under 42 U.S.C. § 1983 for false arrest and false imprisonment. He alleged, inter alia, that defendants’ conduct violated his Fourth Amendment protection from arrest without probable cause and his Fourteenth Amendment right to due process.
The district court bifurcated the trial, separating Grant’s claim against the individual officers from his claim against the City and Police Department. The trial proceeded against the officers (Joseph Ba-hash and Katherine Watson) first and culminated in a jury award of $1.75 million in compensatory and punitive damages in favor of Grant. The officers assign error to the district court’s denial of two motions for judgment as a matter of law, first on probable cause and then on qualified immunity. They also challenge the exclusion of proffered testimony.
We reject the officers’ arguments on all three issues. The district court properly submitted the case to the jury because material issues of fact existed as to the knowledge the officers had at the time of Grant’s arrest and the reasonableness of their actions. The jury was properly in[1084]*1084structed, and there was no error in excluding testimony of questionable relevance.
II. Facts
During the late night hours of September 18, 1998, Carolyn Ronlov was raped by a man she described as white with an ethnic accent. From a scent pad created at the crime scene, a police bloodhound attempted to track the assailant. The dog, named Tinkerbelle, eventually led the officers to a twenty unit apartment building almost two miles away from the crime scene. Tinkerbelle went directly to the second floor of the building, attempted to track the scent for ten more minutes, and then gave up after failing to identify any particular unit or individual. At the time, Grant lived in a unit on the first floor. Tinkerbelle did not show any interest in Grant’s unit or in the first floor at any point.
The officers sought to gather more information by questioning residents awakened by the commotion. Officer Bahash focused his attention on Grant’s particular unit because inside lights appeared to be on but no one answered when he knocked. With Bahash’s permission, two other officers picked the locks on Grant’s apartment door with the intention of drawing the occupant outside. When the door did not open, the police testified that they grew even more suspicious and left with Grant as a possible suspect in their minds. Officer Bahash then obtained a copy of Grant’s photograph from the DMV and placed it in a six-person photograph array (also known as a “six-pack”) for possible identification by two earlier victims. Of the nine victims identified at that point, only Jennifer Haines and Amyjo Dale felt confident enough to make a positive identification from the photo spread.
A. The Haines Identification
On the night of July 3, 1998, Jennifer Haines woke to the sound of someone trying to break into her home. She called 9-1-1 and described the individual as a 5'7" Hispanic man with dark skin tone and short hair. On July 22, approximately two weeks after the incident and before Grant was ever a suspect, Officer Bahash showed Haines a six-pack containing the photograph of a man (Hernandez) that police considered the possible assailant at the time. Haines tentatively identified Hernandez, stating “it looks like number 2, but I’m not real sure. But real close.” A police forensic expert subsequently matched a latent fingerprint found at the Haines residence to one of Hernandez’s rolled fingerprints. The police dropped Hernandez as a suspect shortly thereafter, although neither trial testimony nor the parties’ briefs indicate a reason.
On September 26, almost three months after the attempted break-in, Officer Ba-hash asked Haines to view another six-pack, this time with a photograph of Grant. She selected Grant stating “It’s number 3, if the hair were shorter. It’s him.” Officer Bahash never followed up by having Haines choose between the two photographs of Grant and Hernandez after she had identified both as her assailant.
B. The Dale Identification
On the night of May 25, 1998, Amyjo Dale woke to discover a man crawling on her bedroom floor. They struggled and the assailant succeeded in throwing a blanket over Dale’s head. Just as he was leaving her bedroom, Dale removed the blanket and caught a glimpse of his face. He told her to look away and then fled the scene. She dialed 9-1-1 and described her assailant as a Caucasian male with olive toned skin, about five foot ten to six feet tall. On June 19, three weeks after her attack, Officer Watson showed Dale eight separate photographs. Dale made a tentative identification of an individual named [1085]*1085Oliver, stating his features were very close to her assailant.
On September 30, almost four months after her attack, Officer Watson presented Dale with the array containing Grant’s photograph created by Officer Bahash. At this point, both officers were coordinating their investigation efforts and sharing information. Upon viewing the array, Dale tentatively selected Grant and stated that she was “pretty sure,” but would be more positive with a live identification. On that same day, the police arrested Grant without a warrant for all nine of the Belmont Shores rapes and related felonies.
III. Probable Cause
The officers’ principal argument on appeal is that they had probable cause as a matter of law to arrest Grant without a warrant. This Court reviews de novo the denial of a motion for judgment as a matter of law (JMOL). See Janes v. Wal-Mart Stores Inc., 279 F.3d 883, 886 (9th Cir.2002). “Judgment as a matter of law is appropriate when the evidence, construed in the light most favorable to the nonmoving party, permits only one reasonable conclusion, which is contrary to the jury’s verdict.” Gilbrook v. City of Westminster, 177 F.3d 839, 864 (9th Cir.1999) (emphasis added). If, on the other hand, there is “such relevant evidence as reasonable minds might accept as adequate to support [the jury’s ] conclusion,” then we must affirm the district court’s denial of the officers’ JMOL motion. Id.
Probable cause exists when “under the totality of circumstances known to the arresting officers, a prudent person would have concluded that there was a fair probability that [the defendant] had committed a crime.” United States v. Smith, 790 F.2d 789, 792 (9th Cir.1986). “A police officer has probable cause to effect an arrest if ‘at the moment the .arrest was made ... the facts and circumstances within [his] knowledge and of which [he] had reasonably trustworthy information were sufficient to warrant a prudent man in believing’ that the suspect had violated a criminal law.” Orin v. Barclay, 272 F.3d 1207, 1218 (9th Cir.2001) (quoting Beck v. Ohio, 379 U.S. 89, 91, 85 S.Ct. 223, 13 L.Ed.2d 142 (1964)).
According to testimony presented at trial, Officers Bahash and Watson based their determination of probable cause on the following information: (1) A trained police dog proceeded from a crime scene to Grant’s apartment building in pursuit of a scent trail; (2) Haines and Dale, the only two victims capable of identifying their assailant, selected Grant from a six person photo array; and (3) Grant resembled the general physical description provided by other victims of their assailant. As we discuss below, this evidence does not amount to probable cause at all, much less as a matter of law. The district court properly submitted this issue to the jury and entered judgment upon its verdict.
A. Canine Identification
This Circuit has yet to rale on the extent to which canine identification of an individual may serve as probable cause for an arrest. However, our cases on the use of canines in drag enforcement provide some guidance. For several decades, we have recognized the importance of canine noses to assist in narcotics investigations. See United States v. Solis, 536 F.2d 880, 882 (9th Cir.1976) (“The recent proliferation of crimes involving the transportation of drugs and explosives has led naturally to the training and .use of dogs ... to detect the presence, of such contraband.”). Indeed, we.have routinely held that a canine identification or ‘alert’ of illegal narcotics provides probable cause for the issuance of a search warrant, so long as the dog’s reliability is established. See Unit[1086]*1086ed States v. Lingenfelter, 997 F.2d 632, 639 (9th Cir.1993); see also United States v. Spetz, 721 F.2d 1457, 1464 (9th Cir.1983) (overruled on other grounds) (“A validly-conducted dog sniff can supply the probable cause necessary for issuing a search warrant only if sufficient reliability is established by the application for the warrant.”); United States v. $22,474, 246 F.3d 1212, 1216-17 (9th Cir.2001) (positive alert by police dog with “sophisticated” training constituted probable cause for a search when combined with other corroborating evidence).
We are not alone in placing a heightened reliability standard on canine identifications in narcotics investigations' — other circuits have approached this area with an equally guarded stance. See, e.g., United States v. Ludwig, 10 F.3d 1523, 1527 (10th Cir.1993) (“a dog alert usually is at least as reliable [for the presence of drugs] as many other sources of probable cause” unless “the particular dog had a poor accuracy record”); United States v. Diaz, 25 F.3d 392, 394 (6th Cir.1994) (“For a positive dog reaction to support a determination of probable cause, the training and reliability of the dog must be established.”). Whether a reliable canine identification outside of the drug context provides probable cause for an arrest is an issue we need not decide today. The parties presented enough evidence at trial to call into question the threshold element of Tinkerbelle’s reliability.
Here, Tinkerbelle led police officers two miles from a crime scene to a twenty-unit apartment building full of people at a time when Grant was not at home. After spending ten minutes on the second floor, Tinkerbelle showed signs of confusion and could not identify any particular apartment or individual. She neither showed interest in Grant’s apartment located on the first floor nor any other apartment on that floor. Furthermore, her handler testified that Tinkerbelle was young for a police dog, with only 150 opportunities to track during both training and active duty. Compared to the Lingenfelter dog that had participated in over 500 actual investigations, she was still a novice. The officers did not provide any evidence regarding Tinkerbelle’s accuracy rate to bolster her reliability.
While we recognize the importance of dogs in police investigations, we also adhere to our requirement of reliability as a safeguard against faulty canine identifications. The facts of this case provide no reason to depart from a showing of the dog’s reliability. The jury had good reason to question the reliability of Tinker-belle’s “identification.”
B. Eyewitness Identifications
Whether the identifications supplied by victims Haines and Dale provide probable cause for Grant’s arrest involves two related inquiries: (1) Did the officers employ an identification procedure so impermissibly suggestive as to give rise to a substantial likelihood of misidentification? See Simmons v. United States, 390 U.S. 377, 384, 88 S.Ct. 967, 19 L.Ed.2d 1247 (1968). And if so, (2) did the witnesses exhibit sufficient indicia of reliability to protect the integrity of their identifications? See United States v. Hanigan, 681 F.2d 1127, 1133 (1982).
1. Suggestive Identification Procedures
The Supreme Court has cautioned that “[a] major factor contributing to the high incidence of miscarriage of justice has been the degree of suggestion inherent in the manner in which the prosecution presents the suspect to witnesses for pretrial identification.” United States v. Wade, 388 U.S. 218, 228, 87 S.Ct. 1926, 18 L.Ed.2d 1149 (1967). It has also warned against presenting “pictures of several [1087]*1087persons [where] the photograph of a single individual recurs or is in some way emphasized.” Simmons, 390 U.S. at 383, 88 S.Ct. 967. The officers ignored both of these warnings, Grant contends, when, they placed his photograph next to photographs of five other individuals that neither shared his race nor general facial characteristics.
Race is not an absolute litmus test for the constitutionality of a pretrial identification procedure. See United States v. Portillo, 633 F.2d 1313, 1324 (9th Cir.1980) (upholding photo array that contained only two out of six photographs depicting individuals of Mexican descent). Rather, it is a proxy, along with similarity in face, physique, and dress that courts use to gauge the suggestiveness of a procedure. See, e.g., Raheem v. Kelly, 257 F.3d 122, 137 (2d Cir.2001) (where witnesses described robber as wearing a black leather coat, line up where only one suspect was wearing a black leather coat was impermissibly suggestive); United States v. Barrett, 703 F.2d 1076, 1084-85 (9th Cir.1983) (even though assailant wore sunglasses while committing the crime and defendant was the only person wearing tinted glasses, defendant’s photograph was not so distinct from the others as to render the array unconstitutionally suggestive); Sejnoha v. City of Bisbee, 815 F.Supp. 1300, 1303 (D.Ariz.1993) (photograph array that featured defendant as‘the only Asian person held constitutional because defendant shared similar facial features with at least two other men in the spread); accord Hutsell v. Sayre, 5 F.3d 996, 1005 (6th Cir.1993) (although victim described assailant as young African American male, array containing photographs of six African American males of various ages was not suggestive).
Here, the officers presented Haines and Dale with a photograph array comprised of five Hispanic males and one Caucasian male. The fact that both these racial categories appeared with equal frequency in other victims’ accounts weighs in favor of suggestiveness. At trial, Officer Bahash failed to provide a legitimate reason why he did not create a photograph array with a more equitable distribution of race. Furthermore, Grant’s features bear little resemblance to the others in the array. His face appears long and narrow, whereas four of the other five individuals have rounder, fuller faces. Similarly, Grant’s skin tone appears significantly lighter than four of the five other individuals in the array. That five victims identified their assailant as either Hispanic with light-toned skin or Caucasian with olive-toned skin renders this difference even more salient.
2. Indicia of Reliability
Even though Haines and Dale selected Grant from an arguably suggestive photograph array, their , identifications may still serve as a basis for probable cause if sufficient indicia of reliability are present. Hanigan, 681 F.2d at 1133. In-dicia of reliability include: 1) the opportunity to view the criminal at the time of the crime; 2) the degree of attention paid to the criminal; 3) the accuracy of the prior descriptions of the criminal; 4) the level of •certainty demonstrated at the time of confrontation; and 5) and the length of time between the crime and the confrontation. See Gray v. Klauser, 282 F.3d 633, 639 (9th Cir.2002) (citing Manson v. Brathwaite, 432 U.S. 98, 114, 97 S.Ct. 2243, 53 L.Ed.2d 140 (1977)).
Applying these indicia of reliability, we conclude that the circumstances surrounding the Haines and Dale attacks render their identifications inadequate. Haines viewed her assailant standing outside her window from inside her home for only a few seconds after being startled [1088]*1088awake. See United States v. Field, 625 F.2d 862, 868 (9th Cir.1980) (questioning reliability of identification because witness viewed the robber only twice “both times for only a few seconds and from a distance of 20 feet”). Moreover, she identified Grant almost three months after the attempted break-in and after having tentatively identified another man. Although Dale had a greater opportunity to view her assailant as they struggled in close proximity, the entire encounter lasted only minutes and her head was partially covered by a blanket part of the time. Like Haines, Dale also made a tentative identification of another individual before she selected Grant. See Solomon v. Smith, 645 F.2d 1179, 1186 (2nd Cir.1981) (witness’ identification of a particular defendant becomes less reliable when she identifies other plausible suspects).
In contrast, the circumstances surrounding the Hanigan witness identifications reflected several indicia of reliability. The kidnaping victims had the opportunity to view their captors for several hours during the middle of the day, made their identifications one day after the crime, and were positive in the selection of their captors. 681 F.2d at 1133. Similarly, in Neil v. Biggers, 409 U.S. 188, 199, 93 S.Ct. 375, 34 L.Ed.2d 401 (1972), the Supreme Court held reliable an identification from a suggestive lineup because the victim viewed her rapist for almost 30 minutes under artificial light and moonlight.
C. General Resemblance
The final leg upon which the officers base their claim of probable cause to arrest Grant is perhaps the weakest. They contend that Grant’s resemblance to general physical descriptions provided by earlier victims justified his arrest. Under the law of this Circuit, mere resemblance to a general description is not enough to establish probable cause. See Washington v. Lambert, 98 F.3d 1181, 1190-91 (9th Cir.1996). Not only did the officers have vague descriptions from previous victims of their assailants, the descriptions were often conflicting or incomplete. For instance, one victim described her assailant as 5'10" to 5'11" with lightly tanned skin whereas another victim described her assailant as 5'7" to 5'8" with dark skin. Two victims could not pinpoint their assailant’s race, two others described him as Hispanic, and three others described him as Caucasian. Haines, one of the few victims confident enough to make a positive identification, initially described her assailant to the 9-1-1 operator as 5'7" but then subsequently changed her description to 5'11". Given these vague and disparate descriptions, a 5'7" Hispanic male would have been as likely a suspect as Grant, a Caucasian male at 6'2". See also United States v. Ricardo, 912 F.2d 337, 342 (9th Cir.1990) (officers did not have probable cause to arrest a suspect because he fit the descriptions of a “thin man, not too tall” and “young Mexican male”). As a matter law, Grant’s resemblance to a general physical description could not have provided the officers with probable cause to arrest him.
After considering the totality of the evidence, we hold that the district court properly denied the officers’ JMOL motion. The parties presented enough evidence for a reasonable jury to conclude that Tinker-belle’s “alert” of the wrong apartment building floor, Haines’ and Dale’s arguably unreliable eyewitness identifications, and Grant’s resemblance to vague and conflicting physical descriptions were not enough to establish probable cause for his arrest.
IV. Qualified Immunity
Officers Watson and Bahash also appeal the district court’s denial of their request for qualified immunity as a matter of law. We review this issue de novo. See [1089]*1089Janes, 279 F.3d at 886. When a police officer asserts qualified immunity, we must apply a two-part analysis under Saucier v. Katz, 533 U.S. 194, 121 S.Ct 2151, 150 L.Ed.2d 272 (2001).
The first question is whether the facts, when taken in the light most favorable to Grant, show that the officers’ conduct violated a constitutional right. Id. at 201. Grant’s allegations that the officers arrested him without probable cause satisfy this threshold inquiry. Courts have long held that the Fourth Amendment requires probable cause before an officer may arrest an individual. See Beck v. Ohio, 379 U.S. 89, 91, 85 S.Ct. 223, 13 L.Ed.2d 142 (1964); see also Barlow, 943 F.2d at 1135.
The second question is whether the constitutional right at issue is “clearly established.” Saucier, 533 U.S. at 202, 121 S.Ct. 2151. Here, the “relevant, disposi-tive inquiry in determining whether a right is clearly established is whether it would be clear to a reasonable officer that his conduct was unlawful in the situation he confronted.” Id. Qualified immunity is an objective ' inquiry — whether the officers subjectively believed that they had probable cause to arrest Grant is irrelevant. See Anderson v. Creighton, 483 U.S. 635, 641, 107 S.Ct. 3034, 97 L.Ed.2d 523 (1987). To prevail on their appeal, Officers Watson and Bahash must show as a matter of law that a reasonable officer would’ have arrested Grant based on the canine alert, eyewitness identifications, and his resemblance to a general physical description.
A. The Canine Identification and Related Events
As discussed above, Grant first became a suspect the night a tracking dog led police from the Ronlov crime scene to his apartment building about two miles away. The dog entered the building, went directly to the second floor, and then into the laundry room on the second floor. However, after ten minutes of attempting to pinpoint the scent, the dog became confused. Her handler inferred that it was because the building’s poor ventilation rendered an accurate alert impossible. The dog did not show any interest in sniffing Grant’s apartment at any point throughout the evening.
Grant’s apartment first attracted the officers’ attention because the light was on in the apartment but no one answered the door when police knocked. Officer Bahash testified that he spoke to a woman residing in an adjacent unit who indicated that Grant resembled the general physical description of a composite sketch generated from an earlier victim’s account. Although this anonymous neighbor led Officer Ba-hash to focus more intently on Grant’s apartment, he never mentioned this conversation in his initial or follow up report of that night — this conversation surfaced for the first time after Grant brought his § 1983 claim. The police have subsequently attempted to locate this woman but have failed. Although this Court may . not make credibility judgments about Officer Bahash’s testimony, it was well within the domain of the jury to do so.
At this point, Officer Bahash could have attempted to obtain a search warrant for Grant’s unit. He did not. Instead, he allowed two other officers to pick the top and bottom door locks to Grant’s unit for the purpose of waking the occupant, presumably because the heavy knocking proved ineffective. When the door failed to open after the police picked both locks and applied slight pressure, the officers concluded that it was bolted from the inside and the occupant was hiding from the police.
In essence, because Grant lived in an apartment building identified by a police tracking dog, left the lights on in his unit, and protected his home with locks that were not easily picked, Officers Bahash [1090]*1090and Watson identified him as a suspect for nine rapes and related felonies that occurred in his community. A material issue of fact existed as to whether a reasonable officer would arrive at the same conclusion.
The officers met and interviewed Haines and Dale on several occasions long before they considered Grant a suspect. They knew that the witnesses viewed their assailant under substandard conditions. Both witnesses had only moments to view their assailants and both attacks occurred in the late night hours when the women were asleep. Under these problematic circumstances, the officers could have fortified the identification procedure against possible suggestiveness by showing the victims more than six photographs. See United States v. Sanchez, 24 F.3d 1259, 1262 (10th Cir.1994) (“Common sense dictates that slight irregularities are more likely to ‘jump out’ ... with only six photographs on it than ... a large mug book containing hundreds of photographs.”). Departmental protocol did not require the officers to present a six person photograph array. In fact, when Officer Watson first asked Dale to make an identification back in June 1998, she showed her eight separate photographs. Such a method seems less suggestive because the witness has more photographs to examine and differences among photographs will appear less pronounced.
Alternatively, the officers could have followed the California Peace Officers Legal Source Book, a criminal procedure handbook that was available to them at the time of Grant’s arrest. The Source Book instructs officers to “[ojbtain the photographs of other persons of the same sex and race with similar facial characteristics.” Yet, as we discuss above, they failed to do so. After Haines and Dale selected Grant from the photograph array, the officers could have asked them to compare his photograph to the two individuals they had identified earlier as their possible assailants as an independent verification. They failed to do this as well.
In sum, Officers Bahash and Watson took no steps to protect the identification procedure from suggestiveness nor to verify the accuracy of the identifications once made. A material issue of fact existed as to whether a reasonable officer would have relied on questionable eyewitness identifications without further verification.
Although earlier victims provided conflicting and vague accounts of their assailants, they at least felt confident enough to provide the police with some description to work from. Once Haines and Dale identified Grant, the officers made no attempt to show his photograph to any of these earlier victims to either confirm or deny their descriptions. Even with the pressing need to catch the serial rapist, the officers could have taken the time to call the victim that described her assailant as 5'7" and Hispanic and ask whether Grant, a 6'2" Caucasian make could still fit her description. Whether a reasonable officer would have acted as Officers Watson and Bahash did in this situation was an issue of fact properly decided by the jury.
When viewed in the light most favorable to the non-moving party, there was enough evidence for a reasonable jury to conclude that reasonable officers would not have acted as Officers Bahash and Watson did in arresting Grant. Therefore, the district court properly submitted the issue of qualified immunity to the jury and entered judgment upon its verdict.
V. Evidentiary Rulings
The officers’ final point on appeal is that the district court committed a prej[1091]*1091udicial error in preventing Haines and Dale from making in-court identifications. We review for an abuse of discretion a district court’s rulings on the relevance of evidence. Hanigan, 681 F.2d at 1131. The abuse of discretion standard requires us to uphold a district court determination that falls within a broad range of permissible conclusions in the absence of an erroneous application of law. See Cooter & Gell v. Hartmarx Corp., 496 U.S. 384, 400, 110 S.Ct. 2447, 110 L.Ed.2d 369 (1990). Grant’s civil case pertained to whether the officers had enough information at the time of the arrest to satisfy probable cause — not whether Grant was Haines’ and Dale’s assailant. Therefore, the district court properly refused to admit any evidence that went beyond the scope of what the officers knew at the time of the arrest on September 30, 1998, including any subsequent identifications by victims. Because probable cause and qualified immunity turned on the information that the officers had at the time of Grant’s arrest, the district court reasonably limited the testimony to this subject alone and did not abuse its discretion in preventing Haines and Dale from making in-court identifications of Grant.
The officers also argue that the district court should have allowed testimony by deputy district attorney Goul concerning his participation in the arrest of Grant. At trial, there was evidence that the officers tried to mislead the prosecutor by omitting the following information from their presentation: Haines stated that her assailant was 5'7" on the 9-1-1 tape but subsequently recanted and said 5'11"; Haines previously identified Hernandez as her assailant; Defendant Bahash met with Haines 10-20 times before she identified Grant; Dale previously identified Oliver as her assailant; and Tinkerbelle never showed interest in Grant’s apartment.
Given these omissions, the district court did not abuse its discretion in finding that regardless of Goul’s testimony, no reasonable officer could believe that withholding such information would not have a detrimental effect on a prosecutor’s advice regarding probable cause or qualified immunity.
AFFIRMED.