The problem of judging the effectiveness of a course in general, and of explanations in particular, is certainly one of the most sensible areas in intelligent tutoring systems. In this paper, we present an explanation agent, whose aim is to evaluate the quality of explanations presented to learners. He has two objectives: discovering the source of learner’s misunderstandings by taking into account his student model, and helping the course designer to adapt his explanations according to these observations. We use the conceptual graph theory to structure our explanations into a formal representation. This representation is used by the explanation agent to make his deductions about learners misconceptions