We did a test in my psych class (based on a semi-famous test done years earlier) that really showed just how crappy human memory really is. In this test we're shown "live footage" of a car accident then given a sheet with 10 questions about the accident, all short answer stuff. There were a couple of "seed" questions in there, things phrased to push you towards changing your memory of the accident. Then we collated the answers and rewatched the the accident. My favorite question was "did the pickup truck in the accident have a gun wrack?", only one person in class (about 40, no I wasn't the smart one) answered correctly, which was: what pickup truck, there weren't any. No that's a little unsubtle, but it's lack of subtlety illustrates the point: 39 out of 40 people had their memories edited to include a non-existent pickup because the question made them decide they were wrong.
Keep that in mind when you're on a jury. Eyewitness testimony ain't worth squat. But people worship it.
Interesting footnote--In the parrt I read about line-ups, people were much likelier to make a false ID if there was a suggestion that the perp was in the line-up. It was as though they thought they were "supposed to" find him. When there was no indication one way or the other, their accuracy was nearly 100%. The researchers have tried to educate police departments around the country to improve their ID success and cut down on bad arrests but without much success. Sorry I can't tell you the article, but it may have been in The New Yorker.