Subjects watched a videotape that portrayed a burglar breaking into a home and shooting three innocent victims. The 2-minute tape contained 47 important violent and nonviolent details. Memory for significant details about a violent event improved with each of four successive recall trials, including a trial 48 hours after viewing the tape. Subjects recalled previously unreported details from all segments of the tape on each subsequent trial. These substantial gains in recall occurred without increased guessing and with only a minimal increase in errors. Gains were unaffected by instructions to use context or emotion as retrieval cues. Results suggest that studies of eyewitness memory using a single recall trial may seriously underestimate what a witness can report about events. For example, the present subjects recalled a mean of 38 percent of details on the first trial, but reported 61 percent of details on at least one of the four trials. Results also challenge the assumption that witnesses tell all they know on their first attempt and that later additions to their recollections are less accurate or the result of information provided by investigators. 3 tables and 24 references.