Whether high-precision torque control or motor condition monitoring need accurate motor parameters. For the three parameters of surface-mounted permanent magnet synchronous motor (SPMSM), the voltage equation is rank-deficient. To solve this problem, some scholars proposed methods that build full rank equations with signal injection, but this will produce motor torque ripple, which is not suitable for application to the EPS. Therefore, this paper proposes a method based on MRAS to identify motor parameters step by step. The proposed two steps identification method can make the reference model full rank in every step, but the total decoupling between parameters identification processes cannot be realized for the assumption that the prior step result is the real value. It was found in experiment that this effect varies with the motor operating conditions. Therefore, this paper analyzes the sensitivity of the motor parameters and proposes a method to improve the accuracy of identification results by changing the weight of identification results according to the operating conditions. The performance of the proposed method was verified through experiment, and the accuracy of the estimated results is improved without interfering with the normal operation of the motor.