A sliding mode controller design method based on the synthetically integrated approach is proposed for a nonlinear system. The genetic algorithm is first applied to adjusting the parameters so as to construct an optimized switching function, which ensures the system with better dynamic behavior and enlarged robust range. Then, the method of adjusting the controller parameters online based on neural networks is proposed to overcome the system trajectory's deviating from the switching function due to the uncertainty, and by a learning algorithm with variable learning rate, the convergence of the neural networks is improved. In final, the proposed method is applied to the tension adjusting of the gravure press. The simulation results show the high performance dynamic characteristics and robustness of the proposed controller, as well as the efficiently reduced chattering phenomenon.