The PET H2 15O-bolus method was used to image regional brain activity in normal human subjects during intense pain induced by intradermal injection of capsaicin and during post-capsaicin mechanical allodynia (the perception of pain from a normally non-painful stimulus). Images of regional cerebral blood flow were acquired during six conditions: (i) rest; (ii) light brushing of the forearm; (iii) forearm intradermal injection of capsaicin, (iv) and (v) the waning phases of capsaicin pain; and (vi) allodynia. Allodynia was produced by light brushing adjacent to the capsaicin injection site after ongoing pain from the capsaicin injection had completely subsided. Capsaicin treatment produced activation in many discrete brain regions which we classified as subserving four main functions: sensation-perception (primary somatosensory cortex, thalamus and insula); attention (anterior cingulate cortex); descending pain control (periaqueductal grey); and an extensive network related to sensory-motor integration (supplementary motor cortex, bilateral putamen and insula, anterior lobe and vermis of the cerebellum and superior colliculus). Comparison of the noxious and non-noxious stimuli yielded several new insights into neural organization of pain and tactile sensations. Capsaicin pain, which had no concomitant tactile component, produced little or no activation in secondary somatosensory cortex (SII), whereas light brushing produced a prominent activation of SII, suggesting a differential sensitivity of SII to tactile versus painful stimuli. The cerebellar vermis was strongly activated by capsaicin, whereas light brush and experimental allodynia produced little or no activation, suggesting a selective association with C-fibre stimulation and nociceptive second-order spinal neurons. The experimental allodynia activated a network that partially overlapped those activated by both pain and light brush alone. Unlike capsaicin-induced pain, allodynia was characterized by bilateral activation of inferior prefrontal cortex, suggesting that prefrontal responses to pain are context dependent.