In this paper, the problem of composite anti-disturbance control is investigated for a class of nonlinear system with unknown functions and mismatched disturbances. Firstly, an adaptive neural network disturbance observer is designed to estimate the mismatched disturbances. Secondly, the disturbance estimation is introduced into virtual control law to compensate the mismatched disturbances. In order to avoid “explosion of complexity”, the dynamic surface control technique is employed to achieve the composite controller design. Thirdly, the stability analysis is presented which shows that the controller can guarantee the uniformly ultimate boundedness of the solution of the closed-loop system, and make the tracking error arbitrarily small. Finally, the simulation result is given to demonstrate the effectiveness of the proposed scheme.