The traditional active disturbance rejection control (ADRC) controller in a double closedloop servo system have many parameters and very complicated adjusting processes. In order to solve these problems, an ADRC controller based on a radial basis function neural network is designed. Since the law of the combined control contains the characteristic of independence, a linear state error feedback is designed to further reduce the complexity of parameter setting. The gains of nonlinear errors in an extended stage observer are applied to the radial basis function neural network as weight coefficients, and the online identification for the controlled object’s Jacobian information could be carried out. So, the parameter online selftuning of the ADRC controller could be realized through the selflearning ability of the neural network. Taking a permanent magnet synchronous motor (PMSM) as the controlled object, the experiment is carried out in MATLAB simulation. The results show that this control strategy can effectively optimize the static performance and dynamic quality of the servo system, and the better dynamic and higher precision of the system is reached.