delivered the opinion of the Court.
In this case, we consider whether suggestive behavior by a private party, without any state action, should be evaluated at a pretrial hearing to determine whether an eyewitness’ identification may be admitted at trial.
Here, a husband suspected that his wife had been attacked by his ex-girlfriend. He showed his wife pictures of the woman to help her make an identification, and she then reviewed the photos many times. Afterward, the victim selected a photograph of the ex-girlfriend from a photo array and identified her at trial. Shortly before trial, defendant requested a Wade1 hearing to challenge the admissibility of the identification testimony. The trial judge declined to hold a hearing because no government officer had acted in a suggestive manner.
Recent social science research reveals that suggestive conduct by private actors, as well as government officials, can undermine *311the reliability of eyewitness identifications and inflate witness confidence. We consider that evidence in light of the court’s traditional gatekeeping role to ensure that unreliable, misleading evidence is not presented to jurors. We therefore hold that, even without any police action, when a defendant presents evidence that an identification was made under highly suggestive circumstances that could lead to a mistaken identification, trial judges should conduct a preliminary hearing, upon request, to determine the admissibility of the identification evidence. Accordingly, we remand for an appropriate hearing consistent with the principles outlined below.
I.
On Sunday, January 23, 2005, at 10:16 p.m., Johann Christian Kim (JC) received a phone call from his ex-girlfriend, defendant Cecilia X. Chen. The two had not spoken since June 2000, around the time their relationship ended. JC motioned for his wife, Helen Kim, to come to the phone and listen in on the conversation. During the call, JC relayed that he was happily married and expecting a child. Defendant, by contrast, told him that she was not doing well and had recently broken up with her boyfriend who had mistreated her. She also apologized for having taken JC for granted and wondered aloud “how things might have turned out” had they remained together. According to JC, defendant cried at times before the conversation ended.
Three days later, on January 26, 2005, Helen was home alone recovering from surgery. She was about five months pregnant at the time. While resting, she answered the phone at about 4 p.m., and an unknown woman asked for Mr. Kim. The woman explained that she was calling about a second mortgage with Bank of America, but the Kims had no such mortgage. Helen checked the caller ID after the brief conversation, and it listed “Foley’s Liquor Store”—located in Neptune City. The Kims lived nearby in Ocean Township.
*312Helen went back to sleep but woke up soon after to the sound of loud knocking on the front door. When she responded, she saw a young woman she did not know—whom Helen later identified as the defendant. The woman explained that her car had broken down and asked to use a phone. Helen offered the woman a cell phone but rather than take it, the woman said she needed to use the bathroom. Helen let the woman enter the house and noticed a computer cord sticking out of her left sleeve as the woman walked toward the bathroom.
Helen then called JC to tell him about the strange phone call and the woman. While they were speaking, the woman returned from the bathroom, grabbed Helen, and stabbed her with a kitchen knife in the back of the shoulder. JC heard screams and called 9-1-1.
Helen fought back as the woman continued to try to stab her, at one point cutting Helen’s eyelid. The woman also appeared to try to use the computer cord as a weapon. Helen resisted and grabbed the woman’s wrist. As the struggle continued, the woman said, “I just want your money,” but at no point did she attempt to take anything.
Helen steered the fight toward the front porch and screamed for help. A neighbor who lived across the street, Lori Schoch, heard the screams and saw part of the struggle on the porch. Schoch then called the police.
Eventually, Helen disarmed her attacker. As Helen was about to strike her, the woman said, “Don’t do anything to hurt your baby.” Helen froze in response, and the woman ran off.
When the police responded, Helen described her assailant to Sergeant Thomas Burke as an Asian or Filipino woman, about 5'4", wearing a black jacket, gray hood, dark pants, brown boots, and black gloves. Helen also spoke with Detective Michael Clancy and added that her attacker was about twenty to twenty-five years old, weighed between 110 and 120 pounds, and wore black frame glasses and a gray scarf. (Helen offered a more detailed descrip*313tion late that night at the police station as part of a formal, signed statement.) The police were unable to locate the attacker in the area but recovered a pair of black-rimmed glasses, a steak knife, and a computer cord.
Ms. Schoch provided a similar description to the police within about two hours of the attack. She said the woman appeared to be Asian and was at least 5'6". In a signed statement, Schoch stated, “I didn’t see her face except for a second when she turned____” At trial, some twenty-two months later, Schoch testified that she was able to see the front of the attacker’s face “about ten, fifteen seconds as she turned around,” and the attacker’s profile for approximately thirty seconds.
Helen was unable to sleep after returning from the police station and drew a picture of her assailant. She showed JC the drawing the next morning. He later testified that “it looked a little familiar. Between that and the unusual phone call, I said maybe perhaps it might be [defendant] Cecilia [Chen].” JC then accessed defendant’s personal website and showed Helen five to ten pictures of defendant on the computer. When she saw one particular picture, Helen “just jumped” and was “ninety percent positive” that Chen was her attacker. Helen said she was not completely certain because Chen was smiling and not wearing glasses in the picture, unlike her assailant. In response, Helen’s sister, who was also present, drew eyeglasses onto printed copies of two of the pictures. Helen testified that she looked at the photos about five more times during the first month after the attack.
The Kims brought copies of the photos to the police station later in the day. Helen also worked with a police sketch artist who drew a composite sketch. She later testified that the sketch did not completely resemble her attacker. Schoch saw the sketch in the newspaper, but she, too, thought it was not entirely accurate. On seeing the sketch, though, Schoch realized that she “actually saw more” of the attacker than she had thought.
*314The police conducted an investigation and learned that defendant lived in Maryland and attended medical school there. On the morning of January 26, 2005, defendant worked at a women’s prison in Jessup, Maryland. The police traced her activities that day and determined that she could have left the State in time to arrive in New Jersey and carry out the attack; Other corroborating evidence was also introduced at trial. Among those proofs, a search of defendant’s car uncovered a piece of paper with the words “Ocean Township” and the Kims’ phone number written on it; and a co-worker testified that defendant wore black eyeglasses similar to the glasses found at the Kims’ house.
On April 17, 2006, a grand jury in Monmouth County indicted defendant for the following offenses: two counts of fourth-degree unlawful possession of a weapon, N.J.S.A 2C:39-5d; two counts of third-degree possession of a weapon for an unlawful purpose, N.J.SA 2C:39-4d; aggravated assault, N.J.S.A 2C:12-lb(l); attempted murder, N.J.S.A 2C:5-1, 2C:ll-3; and armed robbery, N.J.SA 2C:15-1.
Nearly twenty-two months after the attack, on November 14, 2006, the police presented a photo array to Helen and Schoch for the first time. Detective Clancy testified that one of the reasons the police waited to show the photo array was out of a “concern” that the website pictures might have prejudiced Helen. Helen selected defendant’s picture; Schoch separately did the same.
Three weeks later, on the eve of trial, defense counsel orally moved for a Wade hearing. Defendant argued that Helen’s identification was based on seeing photos that her husband showed her rather than her memory of the attack. In essence, defendant claimed that Helen identified “a photograph from a photograph.” The trial court denied the motion because “the procedure followed by the police was not impermissibly suggestive.” The court explained that defendant’s arguments went to the weight-of the identifications, not their admissibility.
Trial began the next day on December 5, 2006. Defendant testified on the fourth day of trial. She admitted speaking with *315JC on January 23, 2005, while she was under the influence of alcohol, and confirmed much of his account of the conversation. She denied any role in the attack.
The jury acquitted defendant of robbery and convicted her on all of the other six counts. On March 9, 2007, the court sentenced defendant to a ten-year term of imprisonment for attempted murder, subject to an 85-percent period of parole ineligibility under the No Early Release Act, N.J.S.A 2C:43-7.2, and merged all of the remaining counts into that count of conviction.
Defendant appealed. Preliminarily, the Appellate Division agreed that there was no need to conduct a Wade hearing because law enforcement took no part in the suggestive conduct before the initial identifications. State v. Chen, 402 N.J.Super. 62, 73-77, 952 A.2d 1094 (App.Div.2008). The panel nonetheless observed that identification evidence plays an extremely important role in criminal trials, in that suggestive procedures can contribute to unreliable verdicts and wrongful convictions. Id. at 77, 952 A.2d 1094.
With that in mind, the panel explained that the rules of evidence are designed to root out unreliable evidence. Id. at 77-79, 952 A.2d 1094. As examples, the court noted the need for a pretrial hearing in State v. Michaels, 136 N.J. 299, 316-17, 642 A.2d 1372 (1994), to cleanse the corrupting effects of “coercive or highly suggestive interrogation techniques,” and the bar against presenting hypnotically refreshed testimony to juries because of reliability concerns, in accordance with State v. Moore, 188 N.J. 182, 207, 902 A.2d 1212 (2006). Chen, supra, 402 N.J.Super. at 79-80, 952 A.2d 1094. The panel also observed that other jurisdictions rely on “evidence law to address reliability” in the absence of state action. Id. at 81-82, 952 A.2d 1094 (citations omitted).
Consistent with the courts’ gatekeeping function, the Appellate Division concluded that “trial courts should grant a request for a preliminary hearing when the reliability of the State’s identification evidence is called into question by evidence of highly suggestive words or conduct by private actors that pose a significant risk of misidentification.” Id. at 82, 952 A.2d 1094.
*316The Appellate Division added that when balancing the probative value of identification evidence against the risk of unfair prejudice at a pretrial hearing, trial courts should employ the same analysis used to evaluate suggestive techniques by law enforcement officials. Id. at 83, 952 A.2d 1094 (citing Manson v. Brathwaite, 432 U.S. 98, 114, 97 S.Ct. 2243, 2253, 53 L.Ed.2d 140, 154 (1977)). Ultimately, if the court found “a very substantial likelihood of misidentification,” the evidence would be insufficiently reliable to be admitted under N.J.R.E. 403. Chen, supra, 402 N.J.Super. at 83-84, 952 A.2d 1094 (quoting Neil v. Biggers, 409 U.S. 188, 198, 93 S.Ct. 375, 381, 34 L.Ed.2d 401, 410 (1972)). Otherwise, the reliability of the identification would be for the jury to decide. Id. at 84, 952 A.2d 1094.
Applying that test, the panel concluded that there was sufficient evidence of suggestive conduct to require a prehminary hearing. Id. at 85, 952 A.2d 1094. It therefore remanded the ease to the trial court for a hearing on the admissibility of the identification evidence. Id. at 86, 952 A.2d 1094. If the trial court were to find that the evidence should have been excluded, the panel directed that a new trial be held. Id. at 87, 952 A.2d 1094.
We granted the State’s petition for certification. 197 N.J. 477, 963 A.2d 845 (2009). We also granted the Attorney General’s motion to participate as amicus curiae.
II.
The State argues that the Appellate Division wrongly focused on the reliability of the identification evidence instead of the source of any suggestiveness or taint. It submits that the distinction between public and private conduct should be maintained. In its brief, which preceded the evidentiary hearing in State v. Henderson, 208 N.J. 208, 27 A.3d 872, 2011 WL 3715028 (2011), the State contends that social science studies have focused on identification procedures carried out by law enforcement actors and that private actors lack the same power of suggestion over eyewitnesses.
*317The State concedes that pretrial hearings may be warranted in limited cases: “[ojnly if a defendant can establish ‘grave doubt’ as to the admissibility of identification evidence because the mind of the witness is so clouded by suggestions from non-governmental sources that her personal knowledge is in doubt” (citing N.J.R.E. 602, 701(a)).
Defendant embraces the reasoning in the Appellate Division’s decision. She contends that the opinion sets forth an appropriate test that is supported by the rules of evidence, is consistent with case law in New Jersey and elsewhere, and properly recognizes the effect that private actors can have on the reliability of identification evidence. Defendant therefore agrees that a hearing is warranted on remand.
The Attorney General, appearing as amicus curiae, acknowledges that trial courts have a gatekeeping function under Rules 403 and 104, which may lead to the exclusion of identification evidence influenced by private actors in certain cases. In that regard, the Attorney General proffers the following test: “Only when a defendant can show that highly suggestive conduct by a private actor ... would pose a significant risk of misidentification, such that its probative value is substantially outweighed by its capacity for prejudice, should the identification be excluded.” According to the Attorney General, defendants must carry that burden.
The Attorney General also contends that no remand is needed in this case because the identification was clearly reliable and should have been presented to the jury. If there was any error in not conducting a pretrial hearing, the Attorney General suggests that error was harmless.
III.
A.
This case is not about government conduct. It therefore does not implicate due process concerns raised by suggestive *318police procedures. Cf. Colorado v. Connelly, 479 U.S. 157, 166, 107 S.Ct. 515, 521, 93 L.Ed.2d 473, 483 (1986) (noting that even outrageous behavior by private party seeking to secure evidence against defendant does not make that evidence inadmissible under Due Process Clause); State v. Koedatich, 112 N.J. 225, 316, 548 A.2d 939 (1988) (finding no due process violation when witness identification was not product of suggestive law-enforcement procedures). Here, the victim’s husband is the primary actor, not anyone connected to the police. As a result, we are not concerned about deterring future conduct by law enforcement officers.
Nonetheless, the reasons animating the case law on eyewitness identification extend beyond police procedures and also address the reliability of evidence presented in court. In Manson, swpra, the Supreme Court explained “that reliability is the linchpin in determining the admissibility of identification testimony.” 432 U.S. at 113, 97 S.Ct. at 2253, 53 L.Ed.2d at 154; see also State v. Madison, 109 N.J. 223, 232, 536 A.2d 254 (1988) (describing Manson test as “struggle to balance the State’s need to use eyewitness identification against the defendant’s need to protect himself against potentially unreliable eyewitness testimony”).
We therefore agree with the Appellate Division that although no Wade hearing was necessary, that hardly ends the inquiry. We must consider the admission of eyewitness identifications tainted by private suggestive procedures in light of the rules of evidence and the trial courts’ gatekeeping function.
B.
Courts have a gatekeeping role to ensure that unreliable, misleading evidence is not admitted. Certain basic evidence rules form the bedrock for that principle, as the Appellate Division observed. See Chen, supra, 402 N.J.Super. at 78-79, 952 A.2d 1094. First, only relevant evidence is admissible under Rule 402. That means only evidence “having a tendency in reason to prove or disprove any fact of consequence” may be submitted to a jury. N.J.R.E. 401. Second, even if evidence has some probative value, *319it may be excluded if “the risk of ... undue prejudice, confusion of issues, or misleading the jury” substantially outweighs its probative value. N.J.R.E. 403.
Eyewitness identification testimony, thus, must clear two preliminary hurdles: it must be sufficiently reliable to be able to prove or disprove a fact; and its probative value cannot be substantially outweighed by the risk of undue prejudice or misleading the jury. In addition, eyewitnesses cannot testify unless they have “personal knowledge,” N.J.R.E. 602, and their “opinions and inferences” must be “rationally based on the[ir] perception,” N.J.R.E. 701.
Pieced together, those rules help ensure that certain unreliable evidence is not presented to the jury. They form the trial courts’ gatekeeping function to guarantee that only relevant, probative, and competent evidence that is sufficiently reliable not to run afoul of Rule 403 may be considered by the finder of fact. To carry out that role and determine questions of admissibility, trial judges routinely conduct preliminary hearings under N.J.R.E. 104.
Case law offers a number of examples of how Rule 104 has been used to address reliability concerns generally. In Michaels, supra, for example, this Court ordered a Rule 104 hearing to evaluate whether coercive and unduly suggestive methods used to interrogate children rendered their pretrial statements and testimony so unreliable that they should not be admitted in evidence. 136 N.J. at 315-16, 320, 642 A.2d 1372.
More recently, in State v. A.O., the Court noted that there are sufficient questions about the reliability of polygraph evidence that trial courts must test even stipulated evidence at a Rule 104 hearing before it may be introduced at trial. 198 N.J. 69, 91-92, 965 A.2d 152 (2009).
C.
Against that backdrop, we note that identification evidence has historically raised serious questions about reliability. See, e.g., *320State v. Romero, 191 N.J. 59, 75-76, 922 A.2d 693 (2007) (modifying model jury charge on eyewitness testimony because “eyewitness identification testimony requires close scrutiny and should not be accepted uncritically”); State v. Delgado, 188 N.J. 48, 63-64, 902 A.2d 888 (2006) (requiring police to maintain detailed record of identification procedures); State v. Herrera, 187 N.J. 493, 504, 902 A.2d 177 (2006) (finding showup identification procedures inherently suggestive); State v. Cromedy, 158 N.J. 112, 132, 727 A.2d 457 (1999) (requiring jury instruction on cross-racial identification in certain circumstances).
Today’s decision in Henderson contains a broader examination of the extensive body of scientific evidence that has developed in the past thirty years. Among other things, that evidence reveals the suggestive effect that private actors can have on an eyewitness’ recollection of events. We include the following relevant section from Henderson here for ease of reference:
The current Model Jury Charge states that judges should refer to “factors relating to suggestiveness, that are supported by the evidence,” including “whether the witness was exposed to opinions, descriptions, or identifications given by other witnesses, to photographs or newspaper accounts, or to any other information or influence that may have affected the independence of his/her identification.” Model Jury Charge (Criminal), “Identification: In-Court and Out-of-Court Identifications” (2007). The charge was added after this Court in Herrera invited the Model Jury Charge Committee to consider including express references to suggestibility. Herrera, supra, 187 N.J. at 509-10, 902 A.2d 177 (citing State v. Long, 721 P.2d 483 (Utah 1980)). In response, the Committee relied heavily on proposed charging language in Long.
The Model Jury Charge properly reflects that private—that is, non-State— actors can affect the reliability of eyewitness identifications, just as the police can. The record on remand supports that conclusion. Studies show that witness memories can be altered when co-eyewitnesses share information about what they observed. Those studies bolster the broader finding “that post-identification feedback does not have to be presented by the experimenter or an authoritative figure (e.g. police officer) in order to affect a witness’ subsequent crime-related judgments.” See Blin M. Skagerberg, Co-Witness Feedback in Line-ups, 21 Applied Cognitive Psychol. 489, 494 (2007). Feedback and suggestiveness can come from co-witnesses and others not connected to the State.
Co-witness feedback may cause a person to form a false memoiy of details that he or she never actually observed. In an early study, 200 college students “viewed a film clip, read and evaluated a description of that film ostensibly given by another witness, and wrote out them own description based on their memory of the film.” *321Elizabeth F. Loflus & Edith Greene, Warning: Even Memory for Faces May Be Contagious, 4 Law & Hum. Behav. 323, 328 (1980). The short film depicted a man who parked his car, briefly entered a small grocery store, and upon returning, “got into an argument with a young man who looked as if he were trying to break into the ear.” Ibid.
Some of the students were shown accurate descriptions of the event, and the rest read descriptions that contained false details. See ibid. Some students, for example, observed a young man with straight hair but then read testimony that described the hair as wavy. Id. at 328-29. “This procedure was intended to simulate the situation where a witness to an event is subsequently exposed, either through conversation or reading a newspaper article, to a version given by another witness.” Id. at 324. Results showed that one-third (34%) of students included a false detail—like wavy hair-when they later described the target. Id. at 329. By contrast, only 5% of the students who read a completely factual narrative made similar mistakes. Ibtd. In a related experiment, “[i]f the other witness referred to a misleading detail [a nonexistent mustache], [69]% of the subjects later ‘recognized’ an individual with that feature. Control subjects did so far less often (13%).” Id. at 323, 330.
More recent studies have yielded comparable findings. See Lorraine Hope et al., “With a Little Help from My Friends ...”. The Role of Co-Witness Relationship in Susceptibility to Misinformation, 127 Acta Psychologica 476, 481 (2008) (noting that all participants “were susceptible to misinformation from their co-witness and, as a consequence, produced less accurate recall accounts than participants who did not interact with another witness”); see also Helen M. Paterson & Richard I. Kemp, Comparing Methods of Encountering Post-Event Information: The Power of Co-Witness Suggestion, 20 Applied Cognitive Psychol. 1083, 1083 (2006) (“Results suggest that co-witness information had a particularly strong influence on eyewitness memory, whether encountered through co-witness discussion or indirectly through a third party.”); John S. Shaw, III et al., Co-Witness Information Gan Have Immediate Effects on Eyewitness Memory Reports, 21 Law. & Hum,. Behav. 503, 503, 516 (1997) (“[W]hen participants received incorrect information about a co-witness’s response, they were significantly more likely to give that incorrect response than if they received no co-witness information.”); Rachel Zajae & Nicola Henderson, Don’t It Make My Broum Eyes Blue: Co-Witness Misinformation About a Target’s Appearance Can Impair Target-Absent Lineup Pe'iformance, 17 Memory 266, 275 (2009) (“[Plarticipants who were [wrongly] told by the [co-witness] that the accomplice had blue eyes were significantly more likely than control participants to provide this information when asked to give a verbal description”).
One of the experiments evaluated the effect of the nature of the witnesses’ relationships with one another and compared co-witnesses who were strangers, friends, and couples. Hope et al., supra, at 478. The study found that “witnesses who were previously acquainted with their co-witness (as a friend or romantic partner) were significantly more likely to incorporate information obtained solely from their co-witness into their own accounts.” Id. at 481.
*322Private actors can also affect witness confidence. See [C.A. Elizabeth Luus & Gary L. Wells, The Malleability of Eyewitness Confidence: Co-Witness and, Perseverance Effects, 79 J. Applied Psychol. 714, 714 (1994) ]. In one study, after witnesses made identifications—all of which were incorrect—some witnesses were either told that their co-witness made the same or a different identification. Id. at 717. Confidence rose when witnesses were told that their co-witness agreed with them, and fell when co-witnesses disagreed. See id. at 717-18; see also Skagex-berg, supra, at 494-95 (showing similar results).
In addition, all three experts, Drs. Malpass, Penrod, and Wells, testified at the remand hearing that co-witnesses can influence memory and recall.
To uncover relevant infoi’mation about possible feedback from co-witnesses and other soui'ces, we direct that police officers ask witnesses, as part of the identification process, questions designed to elicit (a) whether the witness has spoken with anyone about the identification and, if so, (b) what was discussed. That information should be recorded and disclosed to defendants. We again rely on our supervisory powers under Article VI, Section 2, Paragraph 3 of the State Constitution in requiring those steps. See Delgado, supra, 188 N.J. at 63, 902 A.2d 888.
Based on the record, we find that non-State actors like co-witnesses and other sources of information can affect the independent nature and reliability of identification evidence and inflate witness confidence—in the same way that law enforcement feedback can. As a result, law enforcement officers should instruct witnesses not to discuss the identification process with fellow witnesses or obtain information from other sources.
[Henderson, supra, 208 N.J. at 268-71, 27 A.3d at 938-40.]
Thus, the State’s argument that the effect of suggestive behavior “is dependent on the [actor’s] status as a police officer” is undermined by reliable scientific findings that are generally accepted by experts in the scientific community. Id. at 247-48, 283, 27 A.3d at 896, 916; see also Rubanick v. Witco Chem. Corp., 125 N.J. 421, 449, 593 A.2d 733 (1991). Even the State, though, concedes that a preliminary hearing is warranted when there is “grave doubt” as to whether a witness’ identification “is so clouded by suggestions from non-governmental sources that her personal knowledge is in doubt.” The Attorney General also accepts that “highly suggestive conduct by a private actor that would pose a significant risk of misidentification” could result in exclusion of identification evidence under Rule 403.
We agree that a private actor’s suggestive words or conduct will require a preliminary hearing under Rule 104 in appropriate cases to test the admissibility of identification evidence. Other courts *323have reached the same conclusion. In State v. McCord, 259 N.J.Super. 217, 218, 611 A.2d 1160 (Law Div.1992), the Law Division evaluated a showup of a suspect conducted by a nongovernmental actor. A private security officer at a shopping mall had detained and handcuffed someone he suspected of a theft. Ibid. A store employee was then summoned to the security office and asked if the person had stolen items from her store. Id. at 219, 611 A.2d 1160. She positively identified the suspect. Ibid.
Absent state action, the trial court recognized that no due process violation had occurred. Id. at 224, 611 A.2d 1160. But it concluded that “basic fairness” required that a pretrial hearing be held when a defendant has made a threshold showing of impermissible suggestiveness. Ibid. The court also elected to apply the traditional two-pronged Manson/Madison test2 at the hearing— even for identification evidence arranged by a civilian. Ibid.
McCord relied on a decision by the Supreme Court of Connecticut in State v. Holliman, 214 Conn. 38, 570 A.2d 680 (1990). In that case, someone raped and robbed a woman walking home from a store at night. Id. at 681. A friend of the victim’s sister recalled that earlier that day, she had been at the same store, and a man had approached her and followed her home. Id. at 682. When the friend spotted the man a few weeks later, she called the *324victim. Ibid. The two then met and pursued the man by ear, and the victim identified him as her attacker. Ibid. The Connecticut Supreme Court agreed with the parties that, even if a defendant’s claim has no constitutional basis, the criteria in Mamón “are appropriate guidelines by which to determine the admissibility of identifications that result from procedures conducted by civilians.” Id. at 684.
More recently, the Supreme Court of Wisconsin resorted to state evidence rules to evaluate the suggestiveness of an accidental, spontaneous identification that did not involve any police conduct. State v. Hibl, 290 Wis.2d 595, 714 N.W.2d 194 (2006). In Hibl, a witness to an accident could only describe the driver to police as a “white male.” Id. at 198. He was unable to provide any other details and did not believe he could identify the individual. Ibid. Seventeen months later, when the witness appeared in court to testify, he saw the defendant leaving the courtroom and positively identified him. Id. at 197-99.
The Hibl court expressed concerns about the reliability of eyewitness identification and observed that factors affecting reliability “do not depend on the presence of a law enforcement procedure.” Id. at 202-03. As a result, the court explained that even when there is no action by law enforcement, courts “still have a gate-keeping function” to assess the reliability of eyewitness identification evidence under Wisconsin’s equivalent of N.J.R.E. 403. Id. at 204-05. In making that determination, the court invited trial judges to consider not only the factors in Mamón but also “the evolving body of law on eyewitness identification” along with “developing scientific research that forms some of its underpinnings.” Id. at 205-06.
Other courts have advanced different theories in concluding that private or accidental suggestive identification procedures warrant pretrial scrutiny. See, e.g., Commonwealth v. Jones, 423 Mass. 99, 666 N.E.2d 994, 1001 (1996) (“Common law principles of fairness dictate that an unreliable identification arising from the especially suggestive [accidental] circumstances of this case should *325not be admitted.”); People v. Blackman, 110 A.D.2d 596, 488 N.Y.S.2d 395, 397 (App.Div.), appeal denied, 65 N.Y.2d 813, 493 N.Y.S.2d 1033, 482 N.E.2d 929 (1985) (remanding for “Wade-type” evidentiary hearing, in case involving allegedly suggestive private action, to ensure fairness and gauge whether identification evidence “meet[s] a threshold of at least minimal reliability”); see also United States v. Bouthot, 878 F.2d 1506, 1513 (1st Cir.1989) (“[Fjederal courts should scrutinize all suggestive identification procedures, not just those orchestrated by the police, to determine if they would sufficiently taint the trial so as to deprive the defendant of due process.”); Green v. Loggins, 614 F.2d 219, 221 (9th Cir.1980) (explaining, in context of accidental jailhouse encounter, that courts review challenged identification “essentially to determine whether the witness’ testimony retains sufficient indicia of reliability”).
To be sure, not all courts have followed the above approach. The State points to United States v. Peele, 574 F.2d 489, 491 (9th Cir.1978), State v. Pailon, 590 A.2d 858, 863 (R.I.1991), and other decisions for the proposition that when government involvement is absent, suggestive identification procedures should be assessed by the jury and not the trial judge. Peele, however, preceded the Ninth Circuit’s decision in Green. In any event, both Peele and Pailón accept that under certain circumstances, highly suggestive behavior that did not involve police action could still result in the exclusion of identification evidence. Peele, supra, 574 F.2d at 491 (“A case might arise where the mind of a witness is so clouded by suggestions from nongovernment sources that a conviction based principally on the testimony of that witness violates due process, but that point was not approached in the instant case.” (citation omitted)); Pailon, supra, 590 A.2d at 863 (“It is conceivable that identification evidence might become so unreliable as to fall below the threshold of competence. This indeed would be a rare occurrence and would involve the question of lack of personal knowledge.” (citations omitted)). Moreover, neither case had the benefit of the record developed in Henderson. We therefore decline to adopt Peele’s high threshold permitting a pretrial hearing only *326when there is “grave doubt” about the admissibility of a witness’ identification testimony. Peele, supra, 574 F.2d at 491. (For the reasons stated in section III.A, we also do not consider cases involving private action as a possible deprivation of due process.)
D.
Because of the pivotal role identification evidence plays in criminal trials, and the risk of misidentification and wrongful conviction from suggestive behavior—whether by governmental or private actors—a private actor’s suggestive words or conduct will require a preliminary hearing under Rule 104 in certain eases to assess whether the identification evidence is admissible. We turn now to consider what the appropriate threshold for a hearing should be in cases that present suggestive identification procedures but no police action.
Today in Henderson, we modified the traditional Manson/Madison test. Specifically, we held that (1) defendants can obtain a pretrial hearing by showing some evidence of suggestiveness that could lead to a mistaken identification, (2) the State must then oiler proof to show that the proffered eyewitness identification is reliable, considering both system and estimator variables, and (3) the ultimate burden is on defendant to prove a very substantial likelihood of irreparable misidentification. Henderson, supra, 208 N.J. at 288-89, 27 A.3d at 919-20. If defendant can make that showing under the totality of the circumstances, the identification evidence is suppressed. Id. at 289, 27 A.3d at 920.
Henderson, like Madison and Manson, addresses the reliability of identification evidence and the need to deter police misconduct. By definition, however, cases that do not involve police action raise no deterrence issues. Simply put, we cannot expect that private actors will conform their behavior to police standards they are unaware of. Absent police involvement, then, our principal concern is reliability.
*327For that reason, we make one modification to Henderson in applying it to cases where there is no police action: we require a higher, initial threshold of suggestiveness to trigger a hearing, namely, some evidence of highly suggestive circumstances as opposed to simply suggestive conduct. That approach devotes extra attention to those cases in which defendants might be able to carry their ultimate burden; if they cannot show highly suggestive private action, it is unlikely they will prevail at the hearing. As a result, the approach also avoids pretrial hearings that would undoubtedly end up being replicated at trial. The value of a hearing in those cases would largely be for deterrence.
Accordingly, we hold that the following modified approach shall apply to assess the admissibility of identification evidence when there is suggestive behavior but no police action: (1) to obtain a pretrial hearing, a defendant must present evidence that the identification was made under highly suggestive circumstances that could lead to a mistaken identification, (2) the State must then offer proof to show that the proffered eyewitness identification is reliable, accounting for system and estimator variables, and (3) defendant has the burden of showing a very substantial likelihood of irreparable misidentification. To reiterate, only the first prong is modified from the test in Henderson. See id. at 288-89, 27 A.3d at 919-20.
At the Rule 104 hearing, courts will weigh both system and estimator variables. Id. at 287-89, 27 A.3d at 918-20. As in Henderson, the court can end the hearing if it finds from the testimony that defendant’s threshold allegation of suggestiveness is groundless. Id. at 289, 290-91, 27 A.3d at 920-21. Ultimately, if the identification evidence is admitted, judges should also make use of enhanced jury instructions at trial and allow expert testimony only if warranted, consistent with Henderson. Id. at 295-98, 27 A.3d at 923-25. We expect that with enhanced jury charges, there will be less of a need for expert witnesses. Id. at 297-98, 27 A.3d at 925.
*328Under the revised formulation, behavior that would trigger a Wade hearing if engaged in by a law enforcement officer would not automatically require a Rule 104 hearing unless the conduct was highly suggestive in its context. Thus, for example, if a police officer conducting a photo array asked, “Are you sure the attacker wasn’t wearing glasses?” that procedure would compel a Rule 104 hearing. The same words uttered in conversation by a friend with no apparent knowledge or authority, though, would not warrant a hearing. By contrast, if an eyewitness provided a detailed identification to a fellow eyewitness, those highly suggestive comments would require exploration at a hearing.
In the end, if a defendant can demonstrate a very substantial likelihood of irreparable misidentification, the identification evidence would not survive scrutiny under Rule 403. Its likelihood to mislead the jury and cause undue prejudice would substantially outweigh any probative value it might offer. In light of the courts’ gatekeeping function, such evidence would properly be excluded under the rules of evidence.
Under the above approach, we recognize—as we did in Henderson—-that in most cases, identification evidence will likely be presented to the jury. Id. at 302-04, 27 A.3d at 928-29. It will remain the jury’s task to determine how reliable that evidence is, with the benefit of cross-examination and appropriate jury instructions. In rare cases, however, highly suggestive procedures that so taint the reliability of a witness’ identification testimony will bar that evidence altogether.
For substantially the same reasons expressed in Henderson, today’s holding applies to defendant Chen and in future cases only. Id. at 300-02, 27 A.3d at 926-28. AlS to others, the ruling will take effect thirty days from the date this Court approves new model jury charges. Id. at 302, 27 A.3d at 927-28.
IV.
Applying the above framework to the facts of this ease, we find that JC’s words and actions were so highly suggestive that a *329pretrial hearing is warranted to assess the admissibility of Helen’s identification evidence. After viewing Helen’s sketch of her assailant, JC not only showed Helen multiple pictures of the defendant from her website, but he also first told Helen the assailant might be “Cecilia.” JC then printed two photos, and defendant was the only person depicted in both of them. Helen also reviewed the selected pictures, with eyeglasses drawn in by hand, at least five more times before trial.
By any standard, the procedure followed was hardly neutral; it strongly suggested that defendant was the assailant. In so finding, we impute no bad faith to JC; we simply recognize that his highly suggestive conduct may have rendered Helen’s identification of defendant unreliable. We also impute no bad faith to the police, who waited twenty-two months before showing Helen a photo array in an attempt to allow any prejudice to dissipate. But we do not suggest that that practice enhanced the reliability of the identification. The delay may not have cleared any taint; plus memories simply do not improve over time. See id. at 266-67, 27 A.3d at 906-07.
We therefore remand this case to the trial court for a Rule 104 hearing consistent with the principles outlined in this opinion. At the hearing, the trial court should evaluate relevant system and estimator variables. In addition to the suggestive nature of the identification procedure, and the possible effect on Helen of multiple viewings, stress, or weapon focus, the trial court should consider Helen’s opportunity to view the attacker before and during their struggle, her attentiveness, and the accuracy of her initial description to the police, among other relevant factors.
We offer no view on the outcome of the hearing. If the trial court finds that the identification should not have been admitted, then a new trial without Helen’s identification evidence is needed. We do not agree with the Attorney General’s suggestion that Helen’s identification testimony was harmless in this case. As the trial judge aptly noted, “[e]veryone knows that a crime was committed ...; identification is the key issue.” Helen *330provided the critical testimony in that regard, notwithstanding other circumstantial, corroborative proofs. If Helen’s identification evidence was properly admitted, then defendant’s conviction should be affirmed.
Helen’s neighbor, Ms. Schoch, identified the defendant as well. Defendant primarily argued before the Appellate Division and this Court that Schoch had insufficient time to observe the attacker.3 Even if true, that claim does not allege that highly suggestive procedures affected Schoch’s testimony.. Accordingly, defendant’s arguments relating to Schoch do not meet the threshold to obtain a hearing.
V.
For the reasons stated above, we modify and affirm the judgment of the Appellate Division and remand the case to the trial court for further proceedings consistent with this opinion.
For affirmance as modified and remandment—Chief Justice RABNER and Justices LONG, LaVECCHIA, RIVERA-SOTO andHOENS—5.
Not Participating—Justice ALBIN.