Part of the success of computerized intelligent tutoring systems will be associated with their ability to assess and diagnose students' knowledge in order to direct pedagogical interventions. What is needed is a methodology for identifying general relationships between on-line action patterns and patterns of knowledge derived off-line. Such a methodology would allow an assessment and diagnosis of knowledge, based only on student actions. The focus of this initial research is the development of a means of identifying meaningful action patterns in student-tutor interactions. Actions executed by subjects on a set of verbal troubleshooting tests (Nichols et al., 1989) were summarized using the Pathfinder network scaling procedure (Schvaneveldt, 1990). The results obtained from this work indicate that meaningful patterns of actions can be identified using the Pathfinder procedure. The network patterns are meaningful in the sense that they can differentiate high and low performers as defined by a previous scoring method. In addition, the networks reveal differences between high and low performers suggestive of targets for intervention.