Neural-musculoskeletal models play a significant role in the interactions between human and robotic devices. Surface Electromyography (sEMG) can effectively measure the electric signal from human muscle and provide useful information for improving the accuracy of human-machine interfaces. This paper summarizes three main sEMG-based research methods at present, establishes the flowchart for sEMGbased musculoskeletal models, and theoretically analyzes the key methods of this interface (which includes sEMG signal filtering, muscle and skeleton model analysis and parameter setting). Also, by using the elbow joint as an example, this paper gathers bicep and tricep signal from experiments, gains muscle activations through Matlab/Simulink software, and simulates joint movement via forward dynamics in OpenSim. By tuning key musculoskeletal parameters, the model's root mean square error (RMSE) for single flexion-extension movement is reduced to 3.98-8.5 degree, showing the feasibility of the potential of using the interface for many applications.