We report a quantitative investigation of three patients who presented with a profound prosopagnosia. They were tested on a series of tests of face perception, face recognition, and paired associate learning using faces. A patient with marked perceptual difficulties but no difficulty whatever in recognizing faces was tested as a control. The control patient's performance was as bad or worse on the tests of perception of faces, and it is concluded that impaired performance on tests of face perception does not contribute to difficulties in recognizing familiar faces. The major difference between the three prosopagnosic patients was the evidence of covert recognition in two of these cases but not in the third. We use this evidence to argue that the perceptual/mnestic distinction is an inappropriate classification and would suggest instead that the two types may be described in terms either of a disconnection of the face recognition units or of damage to the units themselves.