The research focus of sensor-based upper limb motion recognition technology mainly lies in the type and number of sensors, the selection of feature values and the design of classification algorithm. These factors determine the advantages and disadvantages of the recognition effect.Aiming at these problems, this paper proposes an upper limb motion recognition method based on multiple inertial sensors and deep learning.When using the inertial sensor module to collect upper limb movements of the wrist and elbow acceleration and angular velocity data, after the pretreatment extraction such as mean value, maximum (small), frequency of nine kinds of characteristic value, then USES the principal component analysis (pca) to dimension of feature set, in reducing the amount of calculation and at the same time make it has better robustness, finally using deep belief network training data, classify six upper limb movements.Experimental results show that this method can recognize a variety of upper limb movements, and describe complex upper limb movements wit-h the minimum number of sensors. Compared with traditional SVM and ANN algorithms, deep belief network has higher recognition accuracy.