Patients with semantic impairments sometimes demonstrate category-specific deficits suggesting that the anatomical substrates of semantic memory may reflect categorical organisation, however, neuroimaging studies have failed to provide consistent data in support of a category-based account. We conducted three functional neuroimaging experiments to investigate the neural correlates of semantic processing, two with positron emission tomography (PET) and a third with functional magnetic resonance imaging (fMRI). The first experiment used a lexical decision task to search for brain regions selectively activated by concepts from four different categories--animals, fruit, tools, and vehicles. The second experiment used a semantic categorisation task to increase the demands on the semantic system and to look for evidence of consistent activations for the domains of natural kinds or man-made items. The final experiment was a replication of the semantic categorisation task using fMRI to increase the spatial resolution and statistical sensitivity of the experiment. The results of these experiments reliably identified a distributed neural system common to both natural kinds and artifacts but failed to find robust evidence of functional segregation by domain or categories. Category effects were neither reliable nor consistently present across experiments although some were consistent with previous studies. We discuss the implications of these findings, arguing that they are most consistent with a semantic system undifferentiated by category at the neural level.