In this paper, a set of gymnastics motion recognition system based on MEMS inertial sensor is designed to solve the problems of complex background, limited range of activities and personal privacy leakage. The system mainly collects acceleration and angular velocity data of 11 positions when the human body performs gymnastics by constructing a sensor network. Based on the preprocessed two types of data, the parameters such as mean, standard deviation, information entropy and mean square error are calculated as classification features. The support vector machine (SVM) classification model is established and the actions of six gymnastics movements are effectively identified. The experimental results show that the SVM algorithm has better recognition effect than the machine learning algorithms such as Knearest neighbor, naive bayes and decision tree. The average recognition rate can reach over 97%.