We have used neuroimaging techniques, mainly positron emission tomography (PET), to study cognitively driven issues about working memory. Two kinds of experiments are described. In the first kind, we employ standard subtraction logic to uncover the basic components of working memory. These studies indicate that: (a) there are different working-memory systems for spatial, object, and verbal information (with the spatial system localized more in the right hemisphere, and the verbal system more in the left hemisphere); (b) within at least the spatial and verbal systems, separable components seem to be responsible for the passive storage of information and the active maintenance of information (with the storage component being localized more in the back of the brain, and the maintenance component in the front); and (c) there may be separate components responsible for processing the contents of working memory (localized in prefrontal cortex). In our second kind of experiment we have focused on verbal working memory and incrementally varied one task parameter-memory load-in an effort to obtain a more fine-grained analysis of the system's operations. The results indicate that all relevant components of the system show some increase in activity with increasing memory load (e.g., the frontal regions responsible for verbal rehearsal show incremental increases in activation with increasing memory load). In contrast, brain regions that are not part of the working-memory system show no effect of memory load. Furthermore, the time courses of activation may differ for regions that are sensitive to load versus those that are not. Taken together, our results provide support for certain cognitive models of working memory (e.g., Baddeley, 1992) and also suggest some distinctions that these models have not emphasized. And more fundamentally, the results provide a neural base for cognitive models of working memory.