Positron emission tomography (PET) can be used to map brain regions that are active when a visual object (for example, a hand) is discriminated from its mirror form. Chronometric studies suggest that viewers 'solve' this visual shape task by mentally modelling it as a reaching task, implicitly moving their left hand into the orientation of any left-hand stimulus (and conversely for a right-hand stimulus). Here we describe an experiment in which visual and somatic processing are dissociated by presenting right hands to the left visual field and vice versa. Frontal (motor), parietal (somatosensory) and cerebellar (sensorimotor) regions similar to those activated by actual and imagined movement are strongly activated, whereas primary somatosensory and motor cortices are not. We conclude that mental imagery is realized at intermediate-to-high order, modality-specific cortical systems, but does not require primary cortex and is not constrained to the perceptual systems of the presented stimuli.