ONE issue of continuing debate in language research concerns whether the brain holds separate representations for semantic information through the auditoryvsvisual modalities. Regardless of whether we hear, see or read meaningful information, our brains automatically activate both auditory and visual semantic associations to the sensory input. The prominent models for how the brain makes these cross-modality associations holds that semantic information conveyed through either sensory input modality is represented in a shared semantic system comprising the traditionally identified language areas in the brain. A few recent case reports as well as activation imaging studies, have challenged this notion by demonstrating category-specific organization within the semantic system in spatially discrete brain regions. Neither view posits a role for primary sensory cortices in semantic processing. We obtained positron emission tomographic (PET) images while subjects performed an auditory responsive naming task, an auditory analog to visual object naming. Subjects heard and responded to descriptions of concrete objects while blindfolded to prevent visual stimulation. Our results showed that, in addition to traditional language centres, auditory language input produced reciprocal activation in primary and secondary visual brain regions, just as if the language stimuli had entered in the visual modality. These findings provide evidence for a distributed semantic system in which sensory-specific semantic modules are mutually interactive, operating directly onto early sensory processing centers.