Abstract: In order to build a generalized model to relate the electromyography (EMG) signals and the continuous movement variables of the upper limb joints, the motion of human's upper limb is firstly investigated by utilizing the Vicon visual system. Then, the surface EMG (sEMG) signals are sampled from the muscles directly concerned with the upper limb motion, and the muscle activities are extracted from the raw sEMG signals. On the basis, the principal component analysis algorithm is employed to decouple the joint motion and calculate the main elements of muscle activities which determine the joint motion. Finally, a motion model is constructed to map the muscle activities to the multijoint angles of the upper limb via fitting high-order polynomials. Extensive experiments are conducted to verify that the continuous joint angles of the upper limb motion can be accurately estimated by using the proposed model. Moreover, the proposed method is superior to a traditional neural network in estimation performance.