The locations of electromyography (EMG) measurements are directly related to the accuracy of motion recognitionin hand gesture recognition based on EMG signals.This study proposes an EMG measurement position optimization strategy based on ANOVA and back propagation (BP) neural network to obtain the best motion recognition with the fewest EMG sensors.Four EMG sensors are used to capture the EMG signals when the subjects perform specific hand gestures.Feature data extracted from the raw EMG signals are combined into 15 different vectors according to different measurement position combinations.These 15 feature vectors are used to train and test the BP neural network.Single factor analysis of variance (ANOVA)is employed to analyze the significance of the influence of the measured position on themotion recognition.Tukey's honest significant differencetest is adopted to classify the position combinations.The position combinations are divided into several subsets.In the subset with the highest recognition rate,the position combination with the least measurement position and the highest recognition accuracy is considered to be the optimized measurement position.The experimental results show that the measurement position has a significant impact on the results of motion recognition.The accuracy of motion recognition shows an upward trend with the increase in measurement position.The optimal combination of measurement position is P1+P3+P4,and the accuracy of motion recognition is 94.6%.