We used H(2)15O PET to characterize the interaction of words and melody by comparing brain activity measured while subjects spoke or sang the words to a familiar song. Relative increases in activity during speaking vs singing were observed in the left hemisphere, in classical perisylvian language areas including the posterior superior temporal gyrus, supramarginal gyrus, and frontal operculum, as well as in Rolandic cortices and putamen. Relative increases in activity during singing were observed in the right hemisphere: these were maximal in the right anterior superior temporal gyrus and contiguous portions of the insula; relative increases associated with singing were also detected in the right anterior middle temporal gyrus and superior temporal sulcus, medial and dorsolateral prefrontal cortices, mesial temporal cortices and cerebellum, as well as in Rolandic cortices and nucleus accumbens. These results indicate that the production of words in song is associated with activation of regions within right hemisphere areas that are not mirror-image homologues of left hemisphere perisylvian language areas, and suggest that multiple neural networks may be involved in different aspects of singing. Right hemisphere mechanisms may support the fluency-evoking effects of singing in neurological disorders such as stuttering or aphasia.