The correlogram method through a discrete auto regressive (DAR) model is powerful to estimate a damping coefficient and a natural frequency of ship's rolling and pitching motions using the sampled time series of their motions under navigation. But it is difficult to extend the method to a coupled motion, a nonlinear one of ship's oscillation, an input-output motion like maneuvering one and so on. However, nowadays, the development of stochastic identification technique of model is remarkable. In this paper, we propose a new stochastic approach using a continuous auto regressive (CAR) model not only to gain more precise linear parameters of ship's oscillating equation representing roll and pitch under navigation but also to extend the model to the above mentioned more complicated problems in future. The new approach begins with a transform of the equation representing a ship's oscillating motion to a CAR model which will be transformed in the last step to a state space model in accordance with the Kalman's frame work. The likelihood value is calculated by the Kalman filter using information square-root algorithm and the parameters are searched so that the likelihood value takes numerically the maximum value, using the Davidon-Fletcher-Powell optimization technique. The goodness of the model is evaluated by using AIC (Akaike's Informaton Criterion) of the model. The data fitted in this paper are two sets of the roll's and pitch's records which were observed by a container ship under the PNW route in winter. One of them has a single peak and the other one has multiple spectral peaks due to strong rough sea. According to the results, the CAR model proposed here fits well to the actual rolling and pitching data than the conventional method using the correlogram in the viewpoint of the AIC. The estimated natural frequncies of the ship's roll and pitch motion are fairly well but the damping coefficients by a low order's model are not stable especially in the data under rough sea conditions.