In order to enhance the advantage of the autonomous underwater vehicle (AUV), it is necessary to construct a control system which is capable of letting the vehicle swim in the unknown underwater environment. This paper introduces the neural network as a robust and adaptable control system for this purpose. Two procedures with linear feedback control law and Fuzzy control algorithm are carried out to make teaching samples for a cruising type underwater vehicle “PTEROA” through numerical simulations of its motion. Two neural networks are structured by several ten thousand times of updates of their weights with reference to these samples. Both of the networks show good performance in maneuvering a small model of “PTEROA” in a circulating water tunnel in spite of considerable amount of disturbances such as ununiform flow patterns. It should be noted that the original Fuzzy controller failed to control it adequately. It is concluded that the neural network has vast potential for autonomous control systems of underwater vehicles.