Two observers who were using an electronic digital data acquisition system were spot checked for reliability at random times over a four month period. Between-and within-observer reliability was assessed for frequency, duration, and duration-per-event measures of four infant behaviors. The results confirmed the problem of observer drift--the fluctuations of scores across sessions--for the frequency and duration-per-event measures. In contrast, the "real time" duration scores were stable across sessions, indicating the robustness of this measure of behavior. (Author)