Functional magnetic resonance imaging adaptation (fMRIa) is an increasingly popular method that aims to provide insight into the functional properties of subpopulations of neurons within an imaging voxel. The technique relies on the assumption that neural adaptation reduces activity when two successive stimuli activate the same subpopulation but not when they stimulate different subpopulations. Here, we assess the validity of fMRIa by comparing single-cell recordings with functional imaging of orientation, motion and face processing. We find that fMRIa provides novel insight into neural representations in the human brain. However, network responses in general and adaptation in particular are more complex than is often assumed, and an unequivocal interpretation of fMRIa results can be achieved only with great care.