Ten rats were trained on a discrete-trial peak procedure in which food, if available, occurred following the first response after a signal had been present for 10 s. Ten other rats were trained on the same procedure with a 20-s criterion. When the time of reinforcement was changed (10 to 20 s or 20 to 10 s), subjects adjusted their temporal criterion in two abrupt steps. During the intermediate state of the three-step function, rats used a temporal criterion that was near the geometric mean of the initial and terminal times of reinforcement. The conclusion is that the intermediate transition state reflects an internal structure in the animal's information processing system not readily accounted for by stimulus-response models of behavior.