The traditional Extended Kalman Filter (EKF) will lead to large errors of estimation data and robustness weakening of the identification system when applied to brushless DC motor (BLDCM) to estimate rotor position and speed simultaneously. In this paper, based on the discrete mathematical model of BLDCM and M-estimation, an improved EKF(MEKF) was proposed. Firstly, based on the commutation principle of the BLDCM operation and EKF model, a separate commutation model was built up, which was independent to the EKF. Secondly, in order to identify the rotor speed and position with more precision and to enhance the robustness of system, the observation matrix was modified by M-estimation, and a decoupling technology of speed and position was adopted in the system correspondingly. Thirdly, based on the decoupled time series model of the motor, a rotor position detection model was designed out, so that a precision rotor position can be realized in practice without deep filtration which will lead to great lagging. The experiment and simulationresults show that this method can effectively abate the errors disturbance of the EKF, and it also significantly enhances the anti-interference of the initial value and robustness.