We trained rats to perform one of three versions of delayed non-matching-to-sample (DNMS): DNMS between two retractable levers in an enclosed operant chamber; varying-choice DNMS between two arms selected at random on every trial in an uncovered eight-arm radial arm maze; or recurring-choice DNMS between the same two arms on every trial in a covered radial maze (N=33/task). Rats with medial prefrontal cortical lesions showed delay-independent impairments on the retractable lever and recurring-choice tasks, but performed varying-choice DNMS normally. Rats with hippocampal lesions exhibited delay-independent impairments of the retractable lever task and delay-dependent impairments of both radial maze tasks. When rats trained initially to perform recurring choice DNMS were switched to varying choice DNMS, the impairments of both the prefrontal and hippocampal groups were reduced, although hippocampal animals remained significantly impaired. When rats trained initially to perform varying choice DNMS were switched to recurring choice DNMS, the impairment of the hippocampal group was exacerbated while the prefrontal group remained unimpaired. Thus training the prefrontal group to perform the varying choice task first seemed to protect from impairment when these rats were subsequently trained to perform recurring choice DNMS. This protection provides evidence against the possibility that factors related to proactive interference or to temporal discrimination can account for the effects of prefrontal lesions on delayed conditional discriminations involving two response alternatives in fixed locations.