The role of the basal ganglia and cerebellum in the control of movements is unclear. We summarize results from three groups of PET studies of regional CBF. The results show a double dissociation between (i) selection of movements, which induces differential effects in the basal ganglia but not the cerebellum, and (ii) sensory information processing, which involves the cerebellum but not the basal ganglia. The first set of studies concerned motor learning of a sequence of finger movements; there was a shift of activation in the anterior-posterior direction of the basal ganglia which paralleled changes in the motor areas of the frontal cortex. During new learning, the dorsolateral prefrontal cortex and striatum (caudate nucleus and anterior putamen) were activated. When subjects had to select movements, the premotor cortex and mid-putamen were activated. With automatic (overlearned) movements, the sensorimotor cortex and posterior putamen were activated. When subjects paid attention to overlearned actions, activation shifted back to the dorsolateral prefrontal cortex and striatum. The cerebellum was not activated when subjects made new decisions, attended to their actions or selected movements. These results demonstrate components of basal ganglia-(thalamo)-cortical loops in humans. According to earlier studies in animals we propose that the basal ganglia may be concerned with selecting movements or the selection of appropriate muscles to perform a movement selected by cortical areas (e.g. premotor cortex). Secondly, a visuomotor co-ordination task was examined. In the absence of visual control over arm movements, subjects were required to use a computer mouse to either generate new lines or to re-trace lines on a computer screen. The neocerebellum (hemispheres of the posterior lobe, cerebellar nuclei and cerebellar vermis), not the basal ganglia, was more engaged when lines were re-traced (compared with new line generation). Animal experiments have shown that error detection (deviation from given lines) and correction occurs during line re-tracing but not line generation. Our data suggest that the neocerebellum (not the basal ganglia) is involved in monitoring and optimizing movements using sensory (proprioceptive) feedback. Thirdly, the relative contribution of sensory information processing to the signal during active/passive execution of a motor task (flexion and extension of the elbow) was examined; it was found that 80-90% of the neocerebellar signal could be attributed to sensory information processing. The basal ganglia were not involved in sensory information processing. They may be concerned with movement/ muscle selection (efferent motor component); the neocerebellum may be concerned with monitoring the outcome (afferent sensory component) and optimizing movements using sensory (feedback) information.