Using PET, the cerebral network engaged by heard language processing in normal hearing subjects was compared with that in patients who received a cochlear implant after a period of profound deafness. The experimental conditions were words, syllables and environmental sounds, each controlled by a noise baseline. Four categories of effect were observed: (i) regions that were recruited by patients and controls under identical task conditions: the left and right superior temporal cortices and the left insula were activated in both groups in all conditions; (ii) new regions, which were recruited by patients only: the left dorsal occipital cortex showed systematic activation in all conditions versus noise baselines; (iii) regions that were recruited by both groups with a different functional specificity; e.g. Wernicke's area responded specifically to speech sounds in controls but was not specialized in patients; and (iv) regions that were activated in one group more than the other: the precuneus and parahippocampal gyrus (patients more than controls) and the left inferior frontal, left posterior inferior temporal and left and right temporoparietal junction regions (controls more than patients). These data provide evidence for altered functional specificity of the superior temporal cortex, flexible recruitment of brain regions located within and outside the classical language areas and automatic contribution of visual regions to sound recognition in implant patients.