Memory for a staged robbery was tested in two groups of participants witnessing the event either live (n = 62) or on video (n = 64). Immediately after the event participants filled out a questionnaire probing memory with emphasis on the timing of the event and robber characteristics. The results showed that participants who watched a video recording of the event reported more details and with a higher accuracy than participants who were present on the scene, but the pattern of memory errors were similar in the two conditions. It is concluded that laboratory experiments may overestimate the memory of eyewitnesses but are otherwise able to simulate essential aspects of memory performance in naturalistic contexts.