We studied the performance and cortical activation patterns during a mental rotation task (Shepard & Metzler, 1971) using functional magnetic resonance imaging (fMlU) at high field (4 Tesla). Twenty-four human subjects were imaged (fMRI group), whereas six additional subjects performed the task without being imaged (control group). All subjects were shown pairs of perspective drawings of 31, objects and asked to judge whether they were the same or mirror images. The measures of performance examined included (1) the percentage of errors, (2) the speed of performance, calculated as the inverse of the average response time, and (3) the rate of rotation for those object pairs correctly identified as "same." We found the following: (1) Subjects in the fMRI group performed well outside and inside the magnet, and, in the latter case, before and during data acquisition. Moreover, performance over time improved in the same manner as in the control group. These findings indicate that exposure to high magnetic fields does not impair performance in mental rotation. (2) Functional activation data were analyzed from 16 subjects of the fMRI goup. Several cortical areas were activated during task performance. The relations between the measures of performance above and the magnitude of activation of specific cortical areas were investigated by anatomically demarcating these areas of interest and calculating a normalized activation for each one of them. (3) We used the multivariate technique of hierarchical tree modeling to determine functional clustering among areas of interest and performance measures. Two main branches were distinguished: One comprised areas in the right hemisphere and the extrastriate and superior parietal lobules bilaterally, whereas the other comprised areas of the left hemisphere and the frontal pole bilaterally; all three performance measures above clustered with the former branch. Specifically, performance outcome ("percentage of errors") clustered with the parieto-occipital subcluster, whereas both the speed of performance and the rate of mental rotation clustered with the right precentral gyms. We conclude that the mental rotation paradigm used involves the cooperative interaction of functional groups of cortical areas of which some are probably more specifically associated with performance, whereas others may serve a more general function within the task constraints.