When a virtual environment system prevents a user's avatar hand from penetrating virtual objects, the seen and felt positions of the hand separate. We propose a new method for reducing this position discrepancy as quickly as possible without introducing perceptible discrepancy between the seen and felt motion of the user's hand. We performed a user study to compare this new method to two previous methods of dealing with position discrepancy. Our method showed statistically significant improvements in user-reported naturalness and user preference and showed no loss in user performance.