1. Recent concepts concerning animal memory have emphasized the kind of information processed in memory. Reference memory provides information relevant over several trials, i.e., it codes expectancy-based information. Working memory provides information critical for only one trial, i.e., it codes data-based information. Some investigators consider that a continuous alternation task in a T-maze depends on the reference memory of a series of left-right responses, whereas a discrete alternation task is thought to depend on working memory. 2. In the present report, we tested rats in a continuous alternation task with different intertrial intervals (ITI's). Rats were first subjected to 10 or 12 sessions at each of the following ITI's: 0, 55, 100, 200 and 600 s, and then tested at varying ITI's within each session for 12 sessions in the following sequence: 0, 55, 100, 200, 600 and 0 s. Next, the same rats were trained to perform discrete alternation with ITI's and interrun intervals (IRI's) varying across sessions but only IRI's changing within sessions, i.e., IRI = 0 or 55 with ITI = 0, 55, 100, 200 or 600 s across sessions, and IRI = 0, 55, 100, 200 and 600 s with ITI = 0 within sessions. 3. Rats performed both alternation tasks at high levels when ITI's and IRI's changed across sessions. However, when intervals changed within sessions, rats showed a better performance in the continuous task at intervals of 55, 100 and 200 s compared to their performance at these same intervals in the discrete task. In addition, for the discrete task with IRI's changing within sessions, errors probably due to proactive interference occurred more frequently with progressively increasing IRI's. 4. The hypothesis that performance of continuous alternation depends on reference memory and performance of discrete alternation depends on working memory is supported by these data.