In the former report, the author analysed many sea trial's data under the AR autopilot system developed by the authors. Emphasis of the report was mainly put on the evaluation of the course keeping quality. The purpose of this paper is to analyse the same data from the view points of steering law subjecting to the AR autopilot system and the conventional system. One needs to have a robust method not influenced by an existence of feedback loops to attack such a complicated time series related mutually. An answer to this formidable problem is the multi-dimensional auto regressive model decision procedure proposed by Dr. Akaike. The author proposes a new method to analyse actual ship trial data, gained at sea and uses it throughly. He will report in this paper that such an analysis method, especially the noise contribution, the impulse response and the frequency response function calculated from the AR model give important informations to make clear the differences between the optimal steering law, the conventional autopilot steering one and a manual steering one.