Understanding a student's knowledge is an important part of adapting the instruction to that individual student. In the area of Intelligent Tutoring Systems this is called Student Modelling.A specific student modelling problem is studied in the situation where a simulator-based learning environment is used to train process operators in diagnosis. An experiment shows that the information necessary for building a student model is revealed in the dialogue between two students collaborating on diagnosing a fault.As a side effect of this investigation a framework for describing student modelling emerged. In this framework student modelling is viewed as the process of bridging the gap between observations of the student and the system's conception of the knowledge to be taught.This thesis proposes the use of a Learning Companion as the collaboration partner. The ensuing dialogue can then be used for student modelling. An analysis of the collaborative dialogue is presented and several important characteristics identified. An architecture is presented and a prototype that reproduces these characteristics is described.