Surface electromyographic (sEMG) signals can be used in medicine, rehabilitation, robotics and industrial fields. This paper proposes a recognition method of finger motions for dexterous control of prosthetic hands. In this work, we record the corresponding sEMG signals from five subjects, and then set up the model by data preprocessing, feature extraction and classification. The results show that high recognition accuracy can be achieved by using time domain feature extraction and artificial neural network. It means that finger motions can be decoded accurately. To acquire the trade-off between the number of channels and the recognition accuracy, we apply channel number reduction and it is found that the number of channels is at least 7 with recognition accuracy of 90.52%.