This paper studies the factors that influence the accuracy of Virtual Reality (VR) systems, in particular for applications in a Cave Automatic Virtual Environment (CAVE). The CAVE can be used to train surgeon students. For this purpose, an application for a total knee arthroplasty surgery is investigated. To meet the requirements for a high quality training, the accuracy of the tracking system in the CAVE has to be improved. First, a complete model of the tracking system is created based on the extrinsic and intrinsic parameters of the eight-camera system. With this model, the uncertainty of the tracking system is determined for one location in the CAVE. Next, a hybrid method, comprising the Monte Carlo Method and Design Of Experiment, is used to find the most important factors influencing the tracking accuracy in the CAVE.