Much evidence suggests that lesions of the prefrontal cortex (PFC) produce marked impairments in the ability of subjects to shift cognitive set, as exemplified by performance of the Wisconsin Card Sorting Test (WCST). However, studies with humans and experimental primates have suggested that damage to different regions of PFC induce dissociable impairments in two forms of shift learning implicit in the WCST (that is, extradimensional (ED) shift learning and reversal shift learning), with similar deficits also being apparent after damage to basal ganglia structures, especially the caudate nucleus. In this study, we used the same visual discrimination learning paradigm over multidimensional stimuli, and the H215O positron emission tomography (PET) technique, to examine regional cerebral blood flow (rCBF) changes associated with these subcomponent processes of the WCST. In three conditions, subjects were scanned while acquiring visual discriminations involving either (i) the same stimulus dimension as preceding discriminations (intradimensional (ID) shifts); (ii) different stimulus dimensions from previous discriminations (ED shifts) or (iii) reversed stimulus-reward contingencies (reversal shifts). Additionally, subjects were scanned while responding to already learnt discriminations (‘performance baseline’). ED shift learning, relative to ID shift learning, produced activations in prefrontal regions, including, left anterior PFC and right dorsolateral PFC (BA 10 and 9/46). By contrast, reversal learning, relative to ID shift learning, produced activations of the left caudate nucleus. Additionally, compared to reversal and ID shift learning, ED shift learning was associated with relative deactivations in occipito-temporal pathways (for example, BA 17 and 37). These results confirm that, in the context of visual discrimination learning over multidimensional stimuli, the control of an acquired attentional bias or 'set', and the control of previously acquired stimulus-reinforcement associations, activate distinct cortical and subcortical neural stations. Moreover, we propose that the PFC may contribute to the control of attentional-set by modulating attentional processes mediated by occipito-temporal pathways.