Due to non-stationary, modulation, and complex transmission path, the vibrations of gears are more difficult to analysis. Moreover, the parameters of resonance demodulation can not be accurately determined in the traditional envelope analysis for gear vibration analysis. An envelope synchronous average scheme in the angular domain is proposed in the paper for the fault detection of gear. In the proposed approach, the fast kurtogram algorithm is utilized to extract complex envelope signal of gear adaptively at first. Then, the envelope is re-sampled by the even-angle sampling scheme included in the computed order tracking for different reference shafts. Lastly, the order analysis is employed to extract gear fault features after the envelope synchronous average in the angular domain. By this way, broadband noises can be eliminated effectively, the gear fault order components can be extracted from original vibration signal, and the frequency blur caused by speed fluctuation can also be canceled. Simulation and test experiments were conducted respectively to verify the validity of the proposed method.