I respectfully dissent. The majority, despite its holding that the district court committed plain error in this case, affirms Edgar Rene Barrientos-Quintana’s convictions on the ground that the court’s error was harmless. In my view, the court’s failure to instruct the jury that it could not find Barrientos-Quintana guilty based upon the uncorroborated testimony of Marcelo Hernandez affected Barrientos-Quintana’s substantial rights. Accordingly, I would hold that the court’s plain error was not harmless, reverse Barrientos-Quintana’s convictions for first-degree murder and attempted first-degree murder, and remand the case for a new trial.
The majority affirms Barrientos-Quinta-na’s convictions because it concludes that the evidence in the record — specifically, the testimony of two eyewitnesses — corroborates Hernandez’s story that Barrien-tos-Quintana was the shooter. But our precedent requires more of us. In an accomplice-instruction case applying a harmless-error analysis, we have said that “a simple mechanical analysis of the record to determine whether corroborating evidence was presented is not sufficient” to determine harmlessness. State v. Shoop, 441 N.W.2d 475, 480-81 (Minn.1989). We stated that more careful scrutiny is required because
[s]uch an analysis may not disclose the possibility that a jury, in considering the credibility of the witnesses, may have *615rejected the corroborating evidence, leaving the testimony of the accomplice standing alone, uncorroborated. Instead, the focus must be upon the impact of the error on the verdict.
Id. at 481. In addition, “the quantum of corroboration needed” to find an error harmless “varies with the unique circumstances of each case,” and the greater the danger that a particular witness “has been induced to offer incriminating testimony ... based on self-serving motives,” the greater the need for corroboration to “restore confidence.” See State v. Clark, 755 N.W.2d 241, 256 (Minn.2008).
In deciding whether the omission of a required jury instruction likely affected a verdict, we consider the weight of the accomplice testimony relative to other evidence in the record and the overall strength of the evidence corroborating the accomplice’s statements. See Clark, 755 N.W.2d at 255. In State v. Reed, 737 N.W.2d 572, 585 (Minn.2007), for example, we held that even though the failure to give an accomplice instruction was plain error, the error did not affect substantial rights because of the strength of the corroborating evidence. The defendant in Reed was convicted of shooting a police officer. Part of the testimony against Reed came from an accomplice — his girlfriend — who made a false emergency telephone call on Reed’s behalf in order to lure the police officer to the scene of the murder. Id. at 578. But the accomplice’s testimony amounted to a small fraction of the total evidence against Reed. The corroborating evidence formed the primary evidence of Reed’s guilt, including extensive statements by Reed regarding his plan to Mil a police officer and his own admission of guilt to a non-accomplice witness. Id. at 585.
In State v. Clark, a second defendant tried for the murder of the same police officer as in Reed appealed his conviction. 755 N.W.2d at 244-45. We held that the failure to give an accomplice instruction regarding the same accomplice (Reed’s girlfriend) was plain error that affected Clark’s substantial rights and required a new trial. Id. at 252-53. Our analysis in Clark focused on the strength of the corroborating evidence in Reed and the relative paucity of corroboration in Clark’s trial. Id. at 252 (“[T]he State’s evidence at Reed’s trial was stronger than the State’s evidence against Clark.”).
At Clark’s trial, Reed’s girlfriend testified that after she made the emergency call, she and Reed went to meet Clark at his house and stayed there for a brief time — the time during which the officer was shot. Id. at 247. Clark lived just 102 feet from the site where the police officer was Mlled, and his home was situated in the direction from which the fatal bullet was fired. Id. at 254. Witnesses testified about Clark’s association with Reed and the fact that Clark had expressed agreement with Reed’s advocacy for the Mlling of a police officer. Id. Reed and Clark had been seen together many times with a weapon like the one used in the shooting. Id. Further, Clark was seen with Reed, who was carrying a rifle, walMng toward Clark’s house just one-half hour before the shooting. Id. But no witness other than the accomplice (Reed’s girlfriend) placed Clark at the scene of the crime or testified to any statements made by Clark that were suggestive of his involvement. The fact that the accomplice’s testimony was the most important piece of evidence against Clark led us to reverse Clark’s conviction on the ground that the district court failed to give an accomplice instruction.
These recent decisions lead me to conclude that Barrientos-Quintana’s conviction should be reversed for similar *616reasons. Hernandez’s testimony unquestionably formed the heart of the State’s case against Barrientos-Quintana. The only evidence corroborating Hernandez’s testimony came from W.F. and A.B.— two teenage eyewitnesses who were associated with the South Side Raza (SSR) gang that was the target of the shooting. W.F. named “Smokey” as the shooter after first denying any knowledge, and later W.F. made a very tentative photo lineup selection, stating that Barrientos-Quintana “kinda looks like the shooter.” A.B. chose Barrientos-Quintana from a photo lineup. This corroborating testimony was thin evidence of guilt relative to Hernandez’s detailed account of Barrientos-Quintana’s guilt.
If the jury accepted the eyewitnesses’ statements, the statements would amount to “sufficient corroborating evidence ... to restore confidence in the truth of [Hernandez’s] testimony.” See Clark, 755 N.W.2d at 256. The problem, of course, is the not-insignificant possibility that the jurors rejected the corroborating testimony and instead relied on Hernandez’s statements standing alone. See Shoop, 441 N.W.2d at 481. The jurors might have rejected the testimony of W.F. and A.B. for several valid reasons. W.F., A.B., and J.G. (who did not identify Barrientos-Quintana as the shooter) provided inconsistent accounts of the number of people in the vehicle. Their accounts ranged from one to three people in the backseat where the shooter was sitting. W.F. and A.B. were friends of L.P. and his older brother J.P., the leader of the gang Hernandez stated was the target of the shooting. The witnesses admitted that they initially lied to the police about their observations of the shooter, stating that the shooter wore a bandana and sunglasses covering his face. Only W.F. had seen Barrientos-Quintana before, and once presented with photographs to compare with his memory, his selection from the lineup was tentative. Further, the photo of Barrientos-Quintana that was included in the lineup showed him with very close-cut hair — a haircut consistent with all of the eyewitnesses descriptions of the shooter. Photos of Barrientos-Quintana taken on the day of the shooting, by contrast, showed that his hair was much longer than it was in the lineup photo.
Jurors could reasonably have rejected the testimony of the eyewitnesses for any of those reasons and, not knowing that they must find corroboration for the testimony of an accomplice, nonetheless reached a guilty verdict based on Hernandez’s testimony alone. Although the majority correctly notes that the jurors’ request to rehear Hernandez’s testimony does not tell us that the jury in fact rejected the corroborating evidence, that sequence of events does tell us that the jury relied most heavily on Hernandez’s testimony rather than that of the eyewitnesses.
The final relevant consideration in my analysis of the substantial-rights prong is Hernandez’s inherent untrustworthiness as a witness. Cases with highly unreliable accomplices require stronger corroborating evidence than cases like Clark, where we found no basis to conclude that the accomplice testified “based on self-serving motives” but nonetheless reversed the conviction because the corroborating evidence was weak compared to the accomplice’s testimony. Clark, 755 N.W.2d at 256. Hernandez came forth with his story after months of denying any involvement in the shooting, and after twice implicating a third party as the shooter. Only when the police officers implied to Hernandez that he could be treated as a mere “witness” rather than a “suspect” did he implicate Barrientos-Quintana, the person who already had been indicted for the crime. Under the facts as Hernandez describes them, he could have been, but was not, *617charged as an aider and abettor for his role in the shooting.
Moreover, Hernandez was an untrustworthy witness because certain parts of his testimony simply do not hold up when scrutinized. For example, Hernandez claimed that one of Barrientos-Quintana’s motives for the shooting was the fact that Slappy and Beaver told him I.C. was at the • SSR house that evening. Although the evidence showed that I.C. sometimes visited the SSR house, the evidence also unequivocally established that Barrientos-Quintana was at Cub Foods with I.C. about thirty minutes before the murder. I.C. was not at the SSR house on the day of the shooting, and Barrientos-Quintana could not possibly have believed that she was there. That statement about Barrien-tos-Quintana’s purported motive for committing the shooting therefore lacks credibility.
In sum, the record reveals many reasons for the jury to reasonably reject the corroborating evidence. Hernandez was an unreliable witness whose testimony unquestionably formed the heart of the State’s case, and the jury relied heavily on Hernandez’s testimony without knowledge that an accomplice’s testimony must be corroborated. For all of these reasons, I disagree with the majority’s conclusion that the district court’s error in failing to give an accomplice instruction was harmless. I conclude that there is a reasonable likelihood that the court’s error had a significant effect on the verdict. See State v. Griller, 583 N.W.2d 736, 741 (Minn.1998).
Having concluded that Barrientos-Quin-tana has met his burden on the third prong of the plain-error test, I would proceed to the fourth prong. This prong asks whether “reversal ... is necessary ... to ensure fairness and the integrity of the judicial process.” Clark, 755 N.W.2d at 252-53. I conclude that where the district court’s error left the jury unaware of its duty to find corroborating evidence and that error affected the defendant’s substantial rights, the fourth prong is met. Accordingly, I would reverse Barrientos-Quintana’s conviction and remand this case to the district court for a new trial.