An envelope synchronous average scheme for multi-axis gear faults detection is proposed in this paper. In the proposed approach, the vibration from a gearbox is picked up at first. Then, the complex envelope of the vibration is extracted by the fast kurtogram algorithm, which can determine an optimal resonant zone for amplitude demodulation. Subsequently, the envelope is re-sampled by the equal-angular increment sampling scheme to eliminate the possible frequency blur phenomena caused by the speed fluctuation. Moreover, a multi-axis synchronous average strategy is employed to remove the random components in the envelope. Finally, the vibration features of gear faults can be clearly exposed by the envelope order spectra. Compared with the conventional synchronous averaging scheme or the envelope analysis approach, the proposed method is robust for the incipient local fault detection of gears. Experiments are conducted to verify the validity of the proposed method.