Recently PID action feedback control system or optimal control system have been applied to the control of the manoeuvring motion of a ship. But there are some problems for practical use. So it has been tried by authors to apply the feedback-error-learning neural network technique proposed by Kawato et. al. to the automatic manoeuvring system of a ship. The system is called Learning-Feed-Forward Control System. And it has been recognized that the system has a self-tuning ability and a good controllability. In order to push forward with practical use of the control system proposed by authors, there are some items which must be studied. In this paper for the first step of those studies the practical method of the multi-variable control system design is investigated. That is, the real control system must treat multi controlled variables and multi disturbances to a ship motion, and so the method for the decision on many coefficients of the learning equations of neural network and the confirmation of non-interaction between controlled variables are investigated. As a result the practical design method is obtained and it is recognized that the system has a good controllability for the multi-variable control and the multi-diturbance compensation.