Pattern recognition of hand movements based on surface electromyography(sEMG) signals has been widely studied and has good classification performance in healthy subjects. However, for the daily use of amputees, its performance needs to be further studied. In this paper, the electromyography signals of amputees are collected for 10 days to investigate the classification performance of different movements. The practical application conditions are simulated, the time-domain features are taken as the input of Support Vector Machine(SVM), and hand movements recognition is performed on both sides of the arm. As a result, the classification performance of the healthy side is much better than that of the amputation side, because the amputation reduces the stability of the movements. At the same time, different time-domain features are adopted to classify the movements. The results show that the newly proposed feature has better classification performance. The results of classification of different movements show that amputees can control basic hand movements through electromyography signals, but the performance of fine movements needs to be further improved.