In the first experiment, 91 mock witnesses observed a crime, answered questions, made an identification decision, and rated their confidence. It was found that RSA increased the accuracy-confidence correlation for witnesses who made an identification and for those who were high in public self-consciousness. An additional experiment varied accountability and demonstrated that high accountability increased the accuracy-confidence correlation. In this experiment, however, the RSA effect occurred even under low accountability conditions. Observers in both studies could not distinguish between accurate and inaccurate witnesses. With regard to why the RSA effect occurred, there was no support for the hypothesis that videotaped self-exposure serves as a retrieval cue, prompting witnesses to recall the thoughts and feelings that accompanied their prior decisions. Only mixed support was found for the possibility that videotaped self-exposure alerts subjects to informative, but previously unobserved variations in their own overt behavior. Interview data suggested that RSA subjects focused primarily on overt behavioral cues and not on the retrieval of thoughts that accompanied their performance. From a forensic standpoint, the findings corroborated the view that jurors cannot discriminate between accurate and inaccurate witnesses. 50 references and 2 tables (Author abstract modified)