We used positron emission tomography (PET) to answer the following question: Is working memory a unitary storage system, or does it instead include different storage buffers for different kinds of information? In Experiment 1, PET measures were taken while subjects engaged in either a spatial-memory task (retain the position of three dots for 3 sec) or an object-memory task (retain the identity of two objects for 3 sec). The results manifested a striking double dissociation, as the spatial task activated only right-hemisphere regions, whereas the object task activated primarily left-hemisphere regions. The spatial (right-hemisphere) regions included occipital, parietal, and prefrontal areas, while the object (left-hemisphere) regions included inferotemporal and parietal areas. Experiment 2 was similar to Experiment 1 except that the stimuli and trial events were identical for the spatial and object tasks; whether spatial or object memory was required was manipulated by instructions. The PET results once more showed a double dissociation, as the spatial task activated primarily right-hemisphere regions (again including occipital, parietal and prefrontal areas), whereas the object task activated only left-hemisphere regions (again including inferotemporal and parietal areas). Experiment 3 was a strictly behavioral study, which produced another double dissociation. It used the same tasks as Experiment 2, and showed that a variation in spatial similarity affected performance in the spatial but not the object task, whereas a variation in shape similarity affected performance in the object but not the spatial task. Taken together, the results of the three experiments clearly imply that different working-memory buffers are used for storing spatial and object information.