There are three views of face recognition discussed in this paper. The extreme holistic view posits that only the configural information of a face is needed and used in recognition memory tasks. The moderate holistic view understands recognition memory for faces as being reliant upon both configural and featural information but that there is more reliance upon configural information than featural information. Finally, the authors put forward a dual coding theory of face recognition. According to their understanding, both featural information and configural information for faces are encoded and represented in memory and are later used equally on a recognition memory test. 

To test the assumptions made by each of these accounts, Cabeza and Kato constructed configural prototypical faces and featural prototypical faces based upon four other faces. To form a configural prototype of these faces, the shape and skin color information were morphed into a composite. To form a feature prototype for the four faces, the outline and cheeks of the first face, the eyes and eyebrows from the second face, the nose of the third face, and the mouth of the forth face were formed into a single composite face.

Each of the three theories make different predictions about the number of false alarms that will be made by participants for the configural and featural prototypical faces. The extreme holistic view predicts that false alarms should only be made for the configural prototypes. The moderate holistic view predicts that false alarms should be made for both configural and featural prototypical faces, but that more false alarms should be made for configural prototypical faces than for featural prototypical faces. The dual coding theory predicts that an equal number of false alarms should be made for the featural and the configural prototypical faces.

Experiment 1

Materials
288 color photographs of faces were randomly divided into 72 sets of 4 faces. For each of these sets, a featural prototypical face and configural prototypical face was constructed.

Procedure
There were nine sets of faces included in each presentation block. Of these 9 sets of faces, one set included the featural prototypical face, one set included the configural prototypical face and 3 sets were presented without the inclusion of a prototypical face (this only adds up to 5 sets, what are the other 4?).

The test for each of the presentation blocks consisted of:
            one featural and one configural studied prototypical face
            one featural and one configural absent prototypical face
            one featural and one configural new prototypical face
            one studied exemplar
            two new exemplars.

Results
As predicted by the dual coding theory of face recognition, participants gave the same level of confidence of being new to the featural prototypes and the configural prototypes.

Experiment 2

The inversion of faces decreases the ability of individuals to recognize a face. This is often attributed to the inability of people to process the configural information of the face. By inverting the faces, the researchers can setup a situation that will allow them to test the assumption that featural processing does indeed contribute to face recognition. The dual process theory predicts that participants that view inverted faces will false alarm less to configural prototypes than to featural prototypes. This is because inverting the faces will disrupt the encoding of configural information but not the encoding of featural information.

Materials and procedure
30 of the 72 sets of faces were used in the present study. At study participants were presented with the exemplars for 3 sets of faces. At test the participants were presented with the configural and featural prototypes of a set of faces at the same time and were asked to select which of the two had been presented at study. Half of the participants were presented with all upright faces and the other half were presented with all inverted faces.

Results
Participants that viewed the faces upright selected just as many of the configural prototypes as they did the featural prototypes. Participants in that viewed the faces inverted selected more of the featural faces than the configural faces. Reaction times did not yield any significant differences between the conditions.

Discussion
The authors cite the distinction between composite and compromise errors in recollection made by Schooler and Tanaka (1991). After reviewing the literature Schooler and Tanaka concluded that there have been more occurrences of composite recollections than compromise errors. Cabeza and Kato believe the results of these two experiments to be examples of compromise recollections.