This paper presents a control scheme for pneumatic servo system for a practical use, in which a neural network is used to estimate disturbances caused by non-linearity of the system. In the proposed control scheme, the inverse system is directly constructed by the neural network based on an approximated linear model of the plant, which is given as known information. The estimation of the disturbances is realized by subtracting the output signal of the controller from the output signal of the neural network. The effectiveness of the proposed control scheme is confirmed by experiments carried out using the existent pneumatic servo system for practical use. That is, stable output response and excellent positioning accuracy are realized in spite of the substantial non-linearity of the plant.