The flexible-joint manipulator system is a nonlinear high-order system. It is difficult to obtain the kinetic equation accurately. A proposal has been made of an adaptive neural network controller based on the input voltage of the joint drive motor, with its application for the control of the multi-link flexible-joint manipulator system. By adopting the proposed control method, the equations of the joint angle can be obtained by decoupling the flexible joint manipulator model, with motor voltage the inputs of the manipulator system. A neural network controller is designed to approximate the optimal control input, and a robust controller is designed to compensate the approximation error. The control method does not involve complex dynamic equations, thus greatly simplifying the process of calculation. As a result, the proposed control strategy, compared with the existing control methods, is simpler, faster and more efficient. A simulation test has been carried out on the two-link flexible-joint manipulators, thus verifying the effectiveness of the proposed control method.