This paper studies a speed identification method using adaptive unscented Kalman filter (AUKF) for permanent magnet synchronous motor (PMSM) sensorless drive. In AUKF, the adaptive factor can not only adaptively balance the weights of the state equation forecast information and observation information, but also adjust the influence of the system model disturbance on the filtering. Therefore, the stability of sensorless control system for PMSM can be improved. The correctness and the effectiveness of the proposed strategy are tested by experimental results.