Master-slave motion mapping method based on time-varying coefficients is proposed for the dexterous hand driven by ultrasonic motors. In order to evaluate the performance of motion mapping qualitatively, Feature Point Sets(FPSs) of dexterous and human hands are built to describe the whole configurations of both in Cartesian space, and then, a cost function based on generalized Euclidean norm of FPS is established for comparing the similarity between dexterous and human hands. As a result, the time-varying coefficients throughout the whole mapping process are obtained by SQP algorithm. Experimental results show that the proposed method fully meets the requirements of real-time and can achieve good mapping performance.