A Computer-Control method for pneumatic servo system using solenoid switching valves is studied. A learning algorithm based on steepest descent method is proposed for control of the system, in which the detailed information of the system parameters are not needed. The pneumatic system using solenoid switching valves studied in this thesis has advantages of simple construction and low cost compared to a system with servovalves. It is shown that the proposed pneumatic servo system in this study is robust to parameter perturbation, and good position control of the pneumatic cylinder can be achieved by tuning of the velocity feedback gain. The performance of proposed control system is investigated experimentally for the position control of a pneumatic cylinder. Experimental results show that the learning algorithm compensates for the non-linearities of the system, and fast and accurate position control, for which the error remains within ±0.1mm, can be obtained by the use of solenoid switching valves.