Our purpose was to infer the characteristics of the internal clock, temporal memory, and decision processes involved in temporal generalization behavior on the basis of the analysis of individual trials. Three groups of 10 rats each were trained on a peak procedure with reinforcement at 15, 30, or 60 s, with several nonfood trial durations. On nonfood trials, the mean response rate gradually increased to a maximum near the time that reinforcement sometimes occurred and then gradually decreased. Individual trials were characterized by a period of high response rate, preceded and followed by a low response rate. The covariance pattern among measures of the temporal characteristics of the high response rate (start, stop, middle, and spread) supported a parallel, scalar timing model in which animals used on each trial a single sample from memory of the time of reinforcement and separate response thresholds to decide when to start and stop responding. An alternative model, the quasi-serial model (J. Gibbon & R. M. Church, 1992), was not consistent with the obtained relationships between covariances or with the scalar property seen across different nonfood signal durations.