Rats with bilateral neurotoxic reuniens (RE), mediodorsal (MD), hippocampal (HIPP) or sham (SH) lesions were tested in a standard watermaze task, together with unoperated rats. RE-rats and SH-controls readily learned to swim directly to a hidden platform. In contrast, MD-rats displayed a transient deficit characterized initially by thigmotaxis. Like in previous studies, HIPP-rats had long latencies throughout training and displayed more random swims than the other groups. In a memory probe test with the platform removed, SH- and RE-rats approached the correct location relatively directly but, whereas SH-controls persistently searched in the training quadrant, RE-rats switched to searching all over the pool. The MD-group swam in loops to the platform, but then displayed persistent searching in the training quadrant. The HIPP-group performed at chance. These distinct patterns indicate that, although their search strategies were different, RE- and MD-rats had acquired sufficient knowledge about the platform location and could recall information in the probe test. All groups performed well in a subsequent cue test with a visible platform, with RE-rats initially escaping faster than the SH- and HIPP-groups, and MD-rats improving from an initially poorer level of performance to control level. This indicates that there were no sensorimotor or motivational deficits associated with any of the lesions. In conclusion, while the RE and MD nuclei seem not to be critical for the learning and memory of a standard watermaze task, they may contribute to non-mnemonic strategy shifting when animals are challenged in ways that do not occur during training.