Prosopagnosia, a spectacular face agnosia, is generally detected and discussed on the basis of explicit or overt performance of the subjects. However, recent observations suggest that an overt response is probably not sufficient to determine what is and what is not preserved in the cognitive processing of faces by prosopagnosia. Indeed, signs of covert knowledge have been evidenced in some, but not all, subjects, indicating that they are still able to use representations they cannot use overtly. Empirical data, showing that the nature of prosopagnosia is a more complex phenomenon than has been thought and that the processing models should be adjusted accordingly, are reviewed.