We trained 24 rats to perform an eight-arm radial maze task and then assigned them with a matching procedure to one of three treatments: sham surgery or lesions of the projection areas of the mediodorsal thalamic nucleus (MDn) in the medial wall (MW) or in both the MW and rhinal sulcal (RS) areas of frontal cortex. After recovery we trained the rats to perform six tasks, beginning with the standard eight-arm task, followed by two versions of a four forced choice procedure, and then three versions of a two-choice delayed-nonmatching-to-sample (DNMTS) task. The two lesion groups performed comparably on all tasks, showing that impairments were not exacerbated by extension of the MW lesion to include all cortical areas innervated by MDn. As in previous studies, frontal animals performed the radial maze task poorly immediately after surgery but improved with subsequent training. Controlling the order of the arm entries by opening the first four gates in a random sequence had little effect on performance, although frontal animals were impaired when lengthy delays (5 or 15 min) were imposed after the last of the four forced entries. Frontal animals were not impaired on two-choice DNMTS when the arms used for training were selected at random from the eight alternatives on a trial by trial basis, even when visual cues were eliminated by darkening the room and covering the maze. Frontal animals were significantly impaired when the selection of sample and choice arms was limited to the same two alternatives on every trial. This finding may explain the reported sensitivity of DNMTS to the effects of frontal lesions when training is carried out in operant chambers.