This paper, with the linear multivariable systems as a model, proposes a method of evaluating the Kalman gain of the state estimator according to the shifting of poles in the system. The stochasic probability of the process and measurement noise is assumed to be identical to the stationary process. The proposed method, based on the diagonalized system, is easy to implement computationally if the correspondence between the poles of the estimator and the response of the system can be determined. Finally several examples are given to confirm and illustrate proposed method derived here.