A growing number of learning environments on the computer are simulation based. Research shows that such simulations are only effective when the student is guided by a tutor, for instance embedded in an intelligent tutoring system. However, most of the simulations are based on complex mathematical procedures. This makes it hard, if not impossible, for the intelligent tutoring system to use the simulator for explanations or student modelling, because it has no means to relate the calculations within the simulation to the conceptual framework that represents the knowledge the student possesses, or has to acquire. An alternative is to use causal, qualitative models as the basis for simulation. We present an integrated qualitative reasoning approach, implemented as a shell named GARP, that can be used to build both domain models and student models. The modelling language (at the conceptual level) appears to map directly on that of human reasoning about the domain. Furthermore, the resulting models in GARP are very suitable for different kinds of explanations, for diagnosis, and for generating critical tests to induce learning or awareness of misconceptions on the part of the student.