Facial recognition is routine for most people; yet the process of facial recognition or describing a face to be recognized, reveals a great deal of complexity inherent in the activity. Eyewitness identification remains an important element in judicial proceedings: it is very convincing, but it is not very accurate. A study was conducted in which participants were asked to sort a collection of facial photographs into an unrestricted number of piles, based on their individual judgements of facial similarity. Participants then labelled each pile. Stimuli photos were equally divided between two racial groups. These judgements were analyzed to determine if there was any indication of general strategies amongst participants for judging facial similarity. Three potential strategies were identified, based on how the photos were sorted, and how they were labelled. The pile data was converted into binary pairwise similarity ratings. As a result, an information system was obtained with objects corresponding to the participants and attributes corresponding to the pairs of photos. A decision attribute was added for each strategy. Rough set attribute reduction methodology was applied to this data in order to build classifiers for each decision variable. This thesis describes the initial study, the computational approach, the results from an evaluation of the study data, and a list of opportunities for future work. The rough set techniques enable the identification of the sets of photograph pairs that are characteristic of the divisions based on various strategies. Although different participant created quite different sortings of photos, the rough set analysis detected the pairs of photos which seem to have more general significance. These results may lead to a practical test to determine the facial recognition abilities of groups of people, as well as inferring what discriminations people use in face recognition.