The invention relates to an EMG signal recognition method based on wavelet weighted-permutation entropy. Firstly, surface electromyography (sEMG) is collected from related muscle groups of human upperlimbs, action signal segments of the sEMG are determined by the energy threshold, the sEMG of the action signal segments is subjected to four-level decomposition by wavelet transform to obtain wavelet sub-bands of frequency segments, and the weighted-permutation entropy of each wavelet sub-band is calculated; then the calculated weighted-permutation entropies are combined, and the best wavelet sub-band feature combination is selected as the combined feature set; finally, the combined feature set of each wavelet sub-band as a feature vector is input to a support vector machine classifier to realize the hand multi-motion pattern recognition of the EMG signals. The experimental results show that the method can achieve high classification accuracy and better recognition effect when being usedfor hand multi-motion recognition. The method can be effectively applied to pattern recognition of complex daily hand movement, is helpful to improve the flexibility of myoelectric prosthetic hands and makes the myoelectric prosthetic hands better used in life.