The mathematic model for the harmonic drive mechanism with multi- variable and nonlinear external disturbance is established,the system error is defined and virtual function is introduced gradually. The RBF neural network is employed to approximate the nonlinear part of the harmonic drive mechanism on- line.The dynamic surface controller is developed based on RBF neural network. The closed- loop system convergence is demonstrated with the Lyapunov stabilization theory,through the theory analysis and simulation results comparing with general PID control show that the proposed controller not only satisfies high performance position tracking,but also have good accuracy and stability,which effectively suppresses the influence of the harmonic drive mechanism like nonlinear external disturbance.