To develop the guided projectile with higher hit precision, promotes gyro's performance has immensely affection on trajectory measurement. According to the difficulty in method of MEMS angular rate gyro random error on-line modeling, this paper proposes a new algorithm of recursive least-squares method that used in the system identification processing. With this recursive least-squares method, a few known gyro output data are used to roughly estimate system parameters, then newly obtained data are added to the measurement formula using the recursive algorithm in order to make estimated parameters approach to its truth-value. This study indicates that the residual error between outputs of gyro random error model and outputs of real gyro highly approximates white noise, which means the error model accurately describes statistical features of gyro random error and the recursive algorithm realizes on-line modeling.