Teachers gain significant information about their students through close observation of classroom activities. By noting which actions a student takes to achieve particular goals, a teacher can often infer the knowledge possessed by the student and diagnose misconceptions. In this work, we develop a framework for automatically inferring a student’s underlying beliefs from a set of observed actions. This framework relies on modeling how student actions follow from beliefs about the effects of those actions. We demonstrate the practicality of this approach by modeling empirical student data from an educational game and validate its performance via a controlled lab experiment. In the educational game, inferences were consistent with conventional assessment measures; in the lab experiment, the model’s inferences reflect participants’ stated beliefs.