Face perception elicits activation within a distributed cortical network in the human brain. The network includes visual ("core") regions, which process invariant facial features, as well as limbic and prefrontal ("extended") regions that process changeable aspects of faces. Analysis of effective connectivity reveals that the major entry node in the "face network" is the lateral fusiform gyrus and that the functional coupling between the core and the extended systems is content-dependent. A model for face perception is proposed, in which the flow of information through the network is shaped by cognitive demands.