In the first phase, researchers accompanied police officers during interviews with eyewitnesses, and the interview process was reviewed in detail. The second phase involved an examination of police file data to develop profiles of witnesses to various crimes such as homicide, assault, and robbery. During the third phase, interviews were conducted with witnesses, with a focus on crimes having multiple witnesses. Results of this study bear little relationship to the findings of laboratory eyewitness research. Witnesses were more accurate and showed less loss of memory over time than the experimental literature suggests. Few errors in accuracy involved descriptions of actions, and most of these involved witnesses that had viewed the event from an angle that had distorted their view. Descriptions of persons (age, weight, height) were less accurate: witnesses had a 50-50 chance of providing a correct descriptive statistic. Finally, the accuracy of stressed witnesses was significantly better than that of nonstressed eyewitnesses. Directions for conducting future, meaningful field research are discussed, and guidelines are presented to aid police in evaluating proposed research. 4 tables, 1 figure, and 18 references.