Fractional-order modeling for permanent magnet synchronous motor is presented by adopting the combination of mechanism and data in time and frequency domain. The pseudo-random signals is designed to obtain real-time experiment data and numerical fitting to get the fractional order. Then data modeling is realized by using the output-error identification algorithm of the fractional-order system. An enhancement of the classic Levy identification method with weights is applied in the frequency domain. In a real-time permanent magnet synchronous motor plant, the fractional order model is identified according to the experimental tests by using the presented algorithm. The results are compared with two methods of fractional-order modeling, which show the reliability of the results.