In this work, a disturbance-observer based reinforcement learning control scheme is presented for the overhead crane system. First, the approximate/adaptive dynamic programming (ADP) method is applied to obtain the solution of a discounted optimal control problem. Here, we use only one neural network as a critic network. The weights of this network are updated iteratively using a novel updating rule law. A disturbance-observer is then designed to compensate the effect of the unknown input disturbance, therefore improve the robustness of the system. The convergence of each module as well as the stability of the whole closed-loop system is guaranteed by proving rigorously. Finally, numerical simulations are given to illustrate the effectiveness of the proposed method.