Abstract: To realize high-precision position control of the flexible joint driving mechanism, we establish a dynamic model that includes nonlinear factors such as the LuGre friction model, the flexible deformation, and the external disturbance torque. On the basis of this dynamics model, we propose a composite neural dynamic surface controller. The uncertainties of the model are approximated and compensated online by using the RBF neural network (NN). To improve the approximation speed and the accuracy of uncertainties, we construct composite adaptive laws for neural weight updating based on prediction and compensated tracking errors. We guarantee the uniformly ultimate boundedness stability via Lyapunov theory. Compared with the classic dynamic surface control method, the proposed control method is fast and achieves better accuracy in uncertainty approximation, as indicated by simulation results, and the position tracking accuracy of the flexible joint driving mechanism is improved.