Eyewitness identification can be one of the more powerful forms of evidence in trials, but it can also be one of the least accurate. WGBH's Morning Edition Host Joe Mathieu spoke with Northeastern University law professor and WGBH News legal analyst Daniel Medwed about the trouble with eyewitness identification. The transcript below has been edited for clarity.

Joe Mathieu: Today we explore the problem of eyewitness misidentification, which you call possibly the leading cause of wrongful conviction.

Daniel Medwed: That's right. Eight of the first nine documented exonerations of innocent prisoners here in Massachusetts involved an eyewitness misidentification [at trial]. And there's a national study suggesting that in more than 70 percent of documented exonerations, there was an eyewitness misidentification somewhere in the case. Now these aren't situations of a malevolent bad actor intentionally targeting an innocent person, but just a good faith mistake — somebody who just gets it wrong. And we should all be concerned about this, because across the nation, about 75,000 suspects are identified in criminal cases in police precincts.

Mathieu: Why are eyewitnesses wrong so often?

Medwed: Scholars attribute this to a combination of what they call — and pardon the fancy jargon — estimator variables and system variables. First, the estimator variables. This basically relates to the circumstances that surround the initial identification. So most crimes, of course, happen at night. They happened rapidly. They happen unexpectedly, under stressful conditions, which is a perfect storm for creating a mistake. And what's more, human memory is pretty elastic. It's not static. You buy your coffee this morning and you meet the barista, you don't have a concrete, digital recording of that person's face. Instead, your memory shifts in response to events that occur later. So let's say you witness a crime, and then you're walking to the police precinct to report it. You run into a bunch of people who resemble the perpetrator. Those interactions could literally change your image of the perpetrator. In addition, and here's something that I think is critically important, cross-racial misidentifications are particularly problematic. That's where somebody identifies somebody of a different race. And the reason for this is pretty obvious: we still live in a largely segregated society, and we're not intimately familiar with the identifying features of people of different races.

Read More: Hallmarks Of Innocence: The Scope Of Wrongful Convictions

Mathieu: On to system variables, that's what they call the other category here. What would those be?

Medwed: System variables simply relate to police procedures — those procedures that generate eyewitness evidence. So that witness who sees the crime later goes down to the police station to look at one of two things, typically: either what's called a photo array; or a "six pack" — a group of six photographs that loosely resemble the initial description of the perpetrator all shown at once, or a physical lineup of six people in a row. Now a risk with these types of procedures is the witness almost inevitably, naturally is going to think the perpetrator is in there. Why else would the police be asking me to look at these procedures unless they thought the bad guy was there?

Mathieu: So I have to pick one.

Medwed: I have to pick one. And that risk is going to be aggravated by the fact that most police administrators know who the chief suspect is, and they might offer inadvertent cues or clues that would steer the witness toward picking a particular person.

Mathieu: So if eyewitness is so faulty, why do we use it so often?

Medwed: Because it's so powerful. Juries tend to over-rely on them. And to some extent, it's a form of direct evidence that can connect somebody to a particular crime. Most crimes lack biological evidence, they lack physical evidence, so actually eyewitness identification can be a very powerful form of evidence.

Mathieu: Maybe too powerful, in some cases.

Medwed: Maybe too powerful. I think that's right.

Mathieu: So what can we do to prevent misidentifications?

Medwed: Well, there are several potential reforms, and one that a lot of Massachusetts police officials have embraced is called the double-blind lineup, where neither the witness nor the police administrator know who the chief suspect is, or if the chief suspect is even in the lineup. That can greatly improve accuracy.