We report an experiment that assesses the effect of variations in memory load on brain activations that mediate verbal working memory. The paradigm that forms the basis of this experiment is the "n-back" task in which subjects must decide for each letter in a series whether it matches the one presented n items back in the series. This task is of interest because it recruits processes involved in both the storage and manipulation of information in working memory. Variations in task difficulty were accomplished by varying the value of n. As n increased, subjects showed poorer behavioral performance as well as monotonically increasing magnitudes of brain activation in a large number of sites that together have been identified with verbal working-memory processes. By contrast, there was no reliable increase in activation in sites that are unrelated to working memory. These results validate the use of parametric manipulation of task variables in neuroimaging research, and they converge with the subtraction paradigm used most often in neuroimaging. In addition, the data support a model of working memory that includes both storage and executive processes that recruit a network of brain areas, all of which are involved in task performance.