One of the important questions cognitive theories of reasoning must address is whether logical reasoning is inherently sentential or spatial. A sentential model would exploit nonspatial (linguistic) properties of representations whereas a spatial model would exploit spatial properties of representations. In general terms, the linguistic hypothesis predicts that the language processing regions underwrite human reasoning processes, and the spatial hypothesis suggests that the neural structures for perception and motor control contribute the basic representational building blocks used for high-level logical and linguistic reasoning. We carried out a [(15)O] H(2)O PET imaging study to address this issue. Twelve normal volunteers performed three types of deductive reasoning tasks (categorical syllogisms, three-term spatial relational items, and three-term nonspatial relational items) while their regional cerebral blood flow pattern was recorded using [(15)O] H(2)O PET imaging. In the control condition subjects semantically comprehended sets of three sentences. In the deductive reasoning conditions subjects determined whether the third sentence was entailed by the first two sentences. The areas of activation in each reasoning condition were confined to the left hemisphere and were similar to each other and to activation reported in previous studies. They included the left inferior frontal gyrus (Brodmann area 45, 47), a portion of the left middle frontal gyrus (Brodmann area 46), the left middle temporal gyrus (Brodmann areas 21, 22), a region of the left lateral inferior temporal gyrus and superior temporal gyrus (Brodmann areas 22, 37), and a portion of the left cingulate gyrus (Brodmann areas 32, 24). There was no significant right-hemisphere or parietal activation. These results are consistent with previous neuroimaging studies and raise questions about the level of involvement of classic spatial regions in reasoning about linguistically presented spatial relations.