Objective To study the method of hand gesture recognition using surface electromyography ( sEMG) signals from the forearm and hand, as well as joint angles of the thumb and index finger under different hand gestures, so as to explore the feasibility of controlling the exoskeleton hand with sEMG signals. Methods The sEMG signals of six muscles in the right upper limb of 20 healthy right-handed subjects were collected. The time domain feature values of sEMG signals were extracted. Classifiers such as artificial neural network (ANN), Knearest neighbor (KNN), decision tree ( DT), random forest ( RF) and support vector machine (SVM) were used for pattern recognition of six daily hand gestures. Meanwhile, trajectory of the thumb and index finger movements was captured by the Vicon camera tracking system. The thumb and index finger angles were calculated. Results Pattern recognition of six hand gestures could be achieved using sEMG signals of the forearm and hand, and the ANN classifier had the best classification prediction, with test set prediction accuracy of 97. 9% and Kappa coefficient of 0. 975. The thumb and index finger angles under six hand gestures were also calculated, and correlation analysis of joint angles under different hand gestures was conducted. Conclusions By using forearm and hand sEMG signals for hand gesture recognition, it is possible to achieve classification prediction with almost identical results. The results in this study demonstrate the feasibility of applying sEMG signal based-hand gesture recognition to exoskeleton hand control.