To assess dynamic changes in brain function throughout the sleep-wake cycle, CBF was measured with H2(15)O and PET in 37 normal male volunteers: (i) while awake prior to sleep onset; (ii) during Stage 3-4 sleep, i.e. slow wave sleep (SWS); (iii) during rapid eye movement (REM) sleep; and (iv) upon waking following recovery sleep. Subjects were monitored polysomnographically and PET images were acquired throughout the course of a single night. Stage-specific contrasts were performed using statistical parametric mapping. Data were analysed in repeated measures fashion, examining within-subject differences between stages [pre-sleep wakefulness-SWS (n = 20 subjects); SWS-post-sleep wakefulness (n = 14); SWS-REM sleep (n = 7); pre-sleep wakefulness-REM sleep (n = 8); REM sleep-post-sleep wakefulness (n = 7); pre-sleep wakefulness-post-sleep wakefulness (n = 20)]. State dependent changes in the activity of centrencephalic regions, including the brainstem, thalamus and basal forebrain (profound deactivations during SWS and reactivations during REM sleep) are consistent with the idea that these areas are constituents of brain systems which mediate arousal. Shifts in the level of activity of the striatum suggested that the basal ganglia might be more integrally involved in the orchestration of the sleep-wake cycle than previously thought. State-dependent changes in the activity of limbic and paralimbic areas, including the insula, cingulate and mesial temporal cortices, paralleled those observed in centrencephalic structures during both REM sleep and SWS. A functional dissociation between activity in higher order, heteromodal association cortices in the frontal and parietal lobes and unimodal sensory areas of the occipital and temporal lobes appeared to be characteristic of both SWS and REM sleep. SWS was associated with selective deactivation of the heteromodal association areas, while activity in primary and secondary sensory cortices was preserved. SWS may not, as previously thought, represent a generalized decrease in neuronal activity. On the other hand, REM sleep was characterized by selective activation of certain post-rolandic sensory cortices, while activity in the frontoparietal association cortices remained depressed. REM sleep may be characterized by activation of widespread areas of the brain, including the centrencephalic, paralimbic and unimodal sensory regions, with the specific exclusion of areas which normally participate in the highest order analysis and integration of neural information. Deactivation of the heteromodal association areas (the orbital, dorsolateral prefrontal and inferior parietal cortices) constitutes the single feature common to both non-REM and REM sleep states, and may be a defining characteristic of sleep itself. The stages of sleep could also be distinguished by characteristic differences in the relationships between the basal ganglia, thalamic nuclei and neocortical regions of interest.