We studied the performance of young and senior subjects on a well known working memory task, the Operation Span. This is a dual-task in which subjects perform a memory task while simultaneously verifying simple equations. Positron-emission tomography scans were taken during performance. Both young and senior subjects demonstrated a cost in accuracy and latency in the Operation Span compared with performing each component task alone (math verification or memory only). Senior subjects were disproportionately impaired relative to young subjects on the dual-task. When brain activation was examined for senior subjects, we found regions in prefrontal cortex that were active in the dual-task, but not in the component tasks. Similar results were obtained for young subjects who performed relatively poorly on the dual-task; however, for young subjects who performed relatively well in the dual-task, we found no prefrontal regions that were active only in the dual-task. Results are discussed as they relate to the executive component of task switching. W orking memory (WM) is a system for the temporary storage and processing of information. A major reason why researchers are interested in this system is its connection to higher-level cognition. In many computational models of higherlevel cognition that simulate planning, mental calculation, and reasoning, WM is used as a kind of ''mental workspace,'' allowing calculations to be performed on active data structures (1, 2). In behavioral work, there are numerous demonstrations that individual variations in WM capacity, as measured by certain kinds of tasks, are correlated with individual variations in planning and reasoning (e.g., see ref.3). Given the apparent importance of WM to thinking, it is not surprising that there has been a rush of neuroimaging studies in the last several years to explore the neural bases of WM. However, by and large the WM tasks used in neuroimaging are not the kind of tasks that correlate with measures of higher-level cognition. Thus, much of what we know about the neural bases of WM may not bear on the way in which WM is used in higher-level cognition.A WM paradigm that is known to correlate with higher-level cognition, and that we used in our experiment, is the Operation Span (4). On every trial, a sequence of items (e.g., five) is presented, each item consisting of an equation and an unrelated word; the subject has to determine whether the equation is correct and then commit the word to memory, maintaining the words in order. Essentially, this is a dual-task paradigm, involving both math and memory processes. At the end of the trial, memory is tested by a probe that contains the words in a particular order, and subjects have to indicate whether the probe order corresponds to the input order. The microstructure of this task involves a cycle of processing and storage. The subject applies task-specific arithmetic processes to the first equation, then adds a word to WM (a task switch), then processes the second equation while maintaining the WM load (...