BACKGROUND: Mastoidectomy is a complex procedure which can be trained on human cadaveric temporal bones or simulation models. The number of repetitions offered in most training curricula is considerably less than what is normally required for motor skills acquisition in crafts or sports. Directed, self-regulated virtual reality simulation training may provide unlimited repetitions but the effect on learning of extended but unsupervised training is unknown. This study recorded extended learning curves of novices in virtual reality simulation mastoidectomy training. METHODS: Six medical students used the visible ear temporal bone simulator at home for 100 repetitions. Virtual temporal bones were later assessed by 2 blinded experts on a 26-point modified Welling Scale. RESULTS: Four participants completed 100 procedures each and 2 participants completed 50 procedures. Learning curves and plots of time demonstrated great variation: one participant improved gradually during the first 50 procedures and sustained a high performance; another participant achieved only 16 points after 50 procedures; a third participant demonstrated mediocre performances between 10 and 15 points but only used about 5 minutes per procedure. The remaining 3 participants achieved high but fluctuating scores with very limited time use per procedure. Their score per time exceeds the performance of experienced otosurgeons and suggests the use of save/restore gaming strategies to inflate their performance. CONCLUSION: Deliberate learners may reach proficiency in virtual reality simulation of mastoidectomy after 50 repetitions. However, even 100 repetitions cannot guarantee proficiency if motivation fails. Creative “gaming” behavior must be considered and opposed by motivation, supervision, testing, and certification.