The idea of using computer-based surgical simulators for training of prospective surgeons has been a topic of research for more than a decade. However, surgical simulation is still far from being included into the medical curriculum. Still open questions are the level of simulation realism which is needed for effective learning, the identification of surgical skill components which are to be trained, as well as the validation of the training effect. We are striving to address these problems with a new generation of highly realistic simulators. A key element of realism is the variable training scene, reflecting differences in individual patients. In this paper we describe the complete generation process of these case-by-case scenarios.