[15O]-Water positron emission tomography (PET) scanning is an important means of investigating the neural correlates of cognitive processes. This study analyses the statistical power of these experiments under usual experimental conditions. Eleven subjects were scanned repeatedly while they undertook an active-baseline task. Using an empirically derived estimate of variance, the statistical power to detect task-induced increases in regional cerebral blood flow was calculated for different values of magnitude of change, number of subjects and probability of Type I error. The results (a) provide an empirical basis for selecting the number of subjects for cognitive studies using [15O]-water PET; (b) illustrate the importance of considering the statistical power of the test, while making a physiological interpretation of the results; and (c) highlight the problems in concluding the non-involvement of a given brain region in a cognitive task, based solely on the results of a PET study.