A nursery school teacher received three types of training in self-observation skills, which were compared with regard to their effects on the accuracy of the self-observations. Minimal training consisted of a brief description of the behaviors to be recorded. The feedback condition incorporated daily information about the accuracy of the self-observations, and discrimination training involved a detailed training session in which modeling and videotape exercises were used to teach self-observation skills. First, positive verbal behavior was observed, then positive nonverbal behavior. The results led to two major conclusions: (1) The accuracy of a subject's self-observations can be substantially increased through feedback and discrimination training; and (2) the act of self-observing tends to alter the frequency of the behavior being observed. Implications for behavioral self-management programs and teacher training are discussed. (DP)