We have previously developed a prototype virtual-reality-enhanced rehabilitation system using the CyberForce system to assist patients who have suffered from upper extremity stroke to practice some daily life exercises. However, full calibration of the system for each patient is currently not only tedious and time consuming but also impractical in the case of severely disabled hands. In this paper, we propose a practical and easy-to-perform hand-measurement method to calibrate the CyberGlove using artificial neural networks (NNs). The NNs are trained with the hand-segment sizes as input and the manually collected calibrated data as output. The only external device needed is a 2-D digital camera to take the picture of the subject's hand against a chessboard for the hand-segment-size measurement. Subjective evaluation results for various common hand postures show the effectiveness of the proposed method.