Addressing a common problem in the analysis of social networks, this study describes quantitative techniques for identifying social subgroups using individual perceptions of social affinities within natural groups. Compared are four analytic methods for abstracting composite representations of sub-structures. These methods, formally evaluated using Confirmatory Factor Analysis (LISREL IX), are illustrated with a class of 20 seventh-graders enrolled in a regular junior high school. The convergence of the different quantitative techniques towards the same structural representation points to the robustness of the social cognition method for the description of social networks. More than robust, the social cognitive procedure is simple and readily applicable in a wide variety of settings. The approach has several additional positive features. It permits analysis to systematically address the questions of cluster sizes and composition, cluster coherence, and dual memberships. Such comparative analyses are especially useful for tracing developmental changes in the nature of prosocial ties within stable peer groups. (RH)