Prior hypotheses in functional brain imaging are often formulated by constraining the data analysis to regions of interest (ROIs). In this context, this approach yields higher sensitivity than whole brain analyses, which could be particularly important in drug development studies and clinical decision making. Here we systematically examine the effects of different ROI definition criteria on the results inferred from a hypothesis-driven pharmacological fMRI experiment, with the aim of maximising sensitivity and providing a recommended procedure for similar studies. In order to achieve this, we compared different criteria for selecting both anatomical and functional ROIs. Anatomical ROIs were defined (i) specifically for each subject, (ii) at group level, and (iii) using a Talairach-like atlas, in order to assess the effects of inter-subject anatomical variability. Functional ROIs (fROIs) were defined, both for each subject and at group level, by (i) selecting the voxels with the highest Z-score from each study session, and (ii) selecting an inclusive union of significantly active voxels across all sessions. A single value was used to summarise the response within each ROI. For anatomical ROIs we used the mean of the parameter estimates (beta values) of either all voxels or the top 20% active voxels. For fROIs we used the mean beta value of all voxels constituting the ROI. The results were assessed in terms of the achieved sensitivity in detecting the experimental effect. The use of single-subject anatomical ROIs combined with a summary value calculated from the top 20% fraction of active voxels was the most reliable and sensitive approach for detecting the experimental effect. The use of fROIs from individual sessions introduced unacceptable biases in the results, while the use of union fROIs yielded a lower sensitivity than anatomical ROIs. For these reasons, fROIs should be employed with caution when it is not possible to make clear anatomical prior hypotheses.