Abstract: Faults in aero-starter-generator by analyzing the spectrum of motor armature current could be effectively detected and diagnosed. Because rotation speed was fluctuating during the operation of the motor and the ripple of current was proportional to the speed fluctuating, if the sampling period of current was constant, the spectrum of current would spread over a wide range. The fuzziness of spectrum characteristics made fault detection and isolation difficult. While using sampling which was synchronized with the rotation speed this problem could be solved based on analysis of the resulted order ratio spectrum. Experiment on fault detection and isolation was conducted. Four faults (broken coil, broken connection between coil and commutator, short circuit between two neighboring conductors and eccentricity of the rotor) was successively detected and isolated using the proposed method.