It is unclear that we will come to a better understanding of mental processes simply by observing which neural loci are activated while subjects perform a task. Rather, I suggest here that it is better to come armed with a question that directs one to design tasks in ways that take advantage of the strengths of neuroimaging techniques (particularly positron emission tomography and functional magnetic resonance imaging). Here I develop a taxonomy of types of questions that can be easily addressed by such techniques. The first class of questions focuses on how information processing is implemented in the brain; these questions can be posed at a very coarse scale, focusing on the entire system that confers a particular ability, or at increasingly more specific scales, ultimately focusing on individual structures or processes. The second class of questions focuses on specifying when particular processes and structures are invoked; these questions focus on how one can use patterns of activation to infer that specific processes and structures were invoked, and on how processing changes in different circumstances. The use of neuroimaging to address these questions is illustrated with results from experiments on visual cognition, and caveats regarding the logic of inference in each case are noted. Finally, the necessary interplay between neuroimaging and behavioural studies is stressed.