Three experiments used position emission tomography (PET) to study the neural basis of human working memory. These studies ask whether different neural circuits underly verbal and spatial memory. In Experiment 1, subjects had to retain for 3 sec. either the names of four letters (verbal memory) or the positions of three dots (spatial memory). The PET results manifested a clear cut double dissociation, as the verbal task activated primarily left-hemisphere regions whereas the spatial task activated only right-hemisphere regions. In Experiment 2. the identical sequence of letters was presented in all conditions, and what varied was whether subjects had to remember the names of the letters (verbal memory) or their positions in the display (spatial memory). In the verbal task, activation was concentrated more in the left than the right hemisphere; in the spatial task, there was suhstantial activation in both hemispheres, though in key regions, there was more activation in the right than the left hemisphere. Experiment 3 studied only verbal memory, and showed that a continuous memory task activated the same regions as the discrete verbal task used in Experiment 1. Taken together, these results indicate that verbal and spatial working memory are implemented by different neural structures.