The internal torque of fingers is affected by surface electromyography ( sEMG) signals, muscle strength, hand posture and other factors, which cannot be obtained directly. To obtain the torques and coupling torque of finger joints in real time as well as accurately, and apply them to the interactive control of hand rehabilitation robots, a method for analyzing and acquiring the torques and coupling torque of finger joints based on sEMG and musculoskeletal model is proposed. Firstly, an adaptive finger joints angle acquisition system is designed. Through experiments, the sEMG signals of the flexor digitorum superficialis ( FDS) and the extensor digitorum communis (EDC) and the angle data of each finger joint are collected simultaneously. The torque model of the finger joints is formulated and the torque of each joint of the fingers is obtained. Then, the finger D-H model is established, and the coupling torque of the finger joints is obtained by combining the principle of virtual work. Finally, the parameters of the multi-joint torque model are determined, and the simulated torque is achieved through OpenSim software. By comparing the results of the calculated torques and the simulated torques, the root mean square error of three joint torques of the four subjects are 0. 156 7, 0. 097 425 and 0. 084 95, respectively. Results show that the method can obtain the torque and coupling torque of each joint of the finger in real time and accurately, which can meet the requirements of accuracy and real-time interactive control of hand rehabilitation robots.