Without knowledge of how laboratory eyewitnesses behave differently from real eyewitnesses, the relevance and external validity of identification studies may be questioned. The problem of the generalizability of results from laboratory eyewitnesses, however, is met by the problem of the control of variables in real eyewitness studies. Witnesses in identification studies are in social decisionmaking contexts similar to those of real eyewitnesses when their decision to choose someone or to reject the lineup may have a significant impact on others' lives. Two studies presented witnesses with a realistic vandalism situation. In the first, 350 student observers were presented a realistically staged vandalism of moderate seriousness and arousal. This study demonstrated the effects of biased instructions on witnesses' willingness to make a lineup choice and on identification errors (with the offender present and absent). The second study maintained realism through the identification (witnesses were not told that the vandalism had been staged) and showed an unexpected preference of witnesses for making an identification when the supposed consequences for the suspect were to be severe. To evaluate the generalizability of laboratory studies it is important to determine whether their results and related theoretical analyses survive the transposition to more realistic contexts. Realistic studies should serve as benchmarks against which simulations are compared and their generalizability evaluated. Notes, tabular data, and 24 references are provided. (Author abstract modified)