Ten people are secretly filmed as they witness what they believe to be a real crime - a knife attack in a Manchester pub. But when they are later interviewed by the police, their memories are radically different to each other's and to what really happened.

In an extraordinary experiment with the Greater Manchester Police, the problem of eyewitness recollection is dramatically brought into focus, as are the remarkable techniques used by the modern police to counter our unreliable memories.

In the second programme of this series we discover how difficult it is for witnesses to identify suspects.

When bystanders get caught up in a bungled armed robbery and one of them is kidnapped, what effect do weapons and fear have on their memories? High tech eye tracker cameras allow us to compare what witnesses remember from the crime with what they actually saw. The difference between recall and recognition is illustrated by witnesses who can't describe suspects they clearly saw, yet recognise suspects they can't describe.

Having many witnesses to a crime helps the police piece together the facts. But what happens when the only witness is also the victim? Stephanie Slater was blindfolded during her kidnap but her attacker was caught because police interviewers were able to help her recall the sounds and smells she experienced during her ordeal.

But memory can be easily contaminated. Jennifer Thompson-Cannino identified an innocent man as her rapist. He spent 11 years in jail.

Now, British police use cutting-edge techniques designed to collect uncontaminated eyewitness testimony so that they can secure a safe conviction, as in the case of Louise Aird, whose home was broken into during a £2 million art theft.