OBJECTIVE Several training models have been developed to improve surgeons' operative skills as well as patient outcomes. Before implementing these models in training programs, their usefulness and accuracy need to be assessed. In this study, we examined the ability of a laparoscopic nephrectomy (LN) virtual reality (VR) simulator to distinguish between different levels of expertise (construct validity). METHODS Twenty-two novices (no LN experience), 32 intermediates (<10 LN procedures performed) and 10 experienced urologists (> or =10 LN procedures performed) performed the same retroperitoneal task on the LN VR simulator (Mentice, Sweden) three times, performing a practice task before and after the second time. Outcome parameters were time, blood loss, path length, and total score (combination of 62 different parameters). RESULTS No significant differences were found between intermediate and experienced participants. Task 3 performance showed no significant difference between any of the groups. Both intermediates and experienced participants were significantly faster than novices on the first two tasks and had a better total score. Learning curves of intermediate and experienced participants were flat after task two. CONCLUSIONS The LN-VR simulator did not distinguish between intermediate and experienced participants. The analysis of the learning curves suggests that the tasks measured dexterity in using the simulator rather than an actual improvement of operative skills. We conclude that the LN-VR simulator does not have sufficient construct validity and is therefore, in its present form, not suitable for implementation in a urologic training program.