The standard fMRI group analysis is based on statistical detection of task-related brain activity that has a consistent location across brains. The use of this technique has revealed a great deal about the organization of the human brain. Arguably, it has revealed more than has neuropsychology and has done so in a much shorter time. Such studies continue to proliferate but, I shall argue, they have inherent limitations that severely reduce the life expectancy of the approach. One practical limitation is simply that averaging across brains (even though spatially normalised) discards much of the spatial precision that will form the bedrock of fMRI in the future. A more fundamental limitation is that, as has often been pointed out by critics, knowing where something occurs is not the same as knowing how it works. I shall review some promising avenues down which fMRI research may be able to move in order to overcome the limitations of the standard group analysis. I shall then illustrate one of them (the repetition suppression paradigm) with my own work on the processing of optic flow in the occipital cortex.