The perception and memory of faces have been accounted for by the processing of two kinds of facial information: featural and configurational. The starting point of this article is the definition and accepted usage of these two concepts of facial information. I discuss these definitions and their various ramifications from three aspects: methodological, theoretical and empirical. In the section on methodology, I review several of the basic manipulations for changing facial information. In the theoretical section, I consider four fundamental hypotheses associated with these two kinds of facial information: the featural, configurational, holistic and norm hypotheses (the norm-based hypothesis and the ‘hierarchy of schemas’ hypothesis). In the section on empirical evidence, I survey relevant studies on the topic and consider these hypotheses through a description of various empirical phenomena that carry clear implications for the subject of the study. In conclusion, I propose two alternative directions for future research: first, a ‘task-information’ approach, which involves specifying what information is used for different tasks; and secondly, taking a different approach to the definition of the visual features for face processing, for example by using principal components analysis (PCA).