Background: Training within a proficiency‐based virtual reality (VR) curriculum may reduce errors during real surgical procedures. This study used a scientific methodology for development of a VR training curriculum for laparoscopic cholecystectomy. Methods: Inexperienced (had performed fewer than ten laparoscopic cholecystectomies), intermediate (20–50) and experienced (more than 100) surgeons were recruited. Construct validity was defined as the ability to differentiate between the three levels of experience, based on simulator‐derived metrics for nine basic skills, four procedural tasks and full laparoscopic cholecystectomy on a high‐fidelity VR simulator. Inexperienced subjects performed ten repetitions for learning curve analysis. Proficiency measures were based on the performance of experienced surgeons. Results: Thirty inexperienced, 11 intermediate and 16 experienced operators were recruited. Eight of nine basic skills and three of four procedural tasks were found to be construct valid. The full procedure revealed significant intergroup differences for time (1541, 673 and 816 s; P = 0·002), movements (1021, 595 and 638; P = 0·006) and path length (2038, 1235 and 1303 cm; P = 0·033). Learning curves plateaued between the second and ninth sessions. Conclusion: This study shows that it is possible to define and develop a whole‐procedure VR training curriculum for laparoscopic cholecystectomy using structured scientific methodology. Copyright © 2009 British Journal of Surgery Society Ltd. Published by John Wiley & Sons, Ltd. Simulation‐based training