In order to monitor and maintain turbine gearbox in time, This article presented a diagnosing method to turbine gearbox crack based on wavelet packet and cepstrum analysis. According to the characteristics that gear crack vibration signals being speed-frequency modulating meshing frequency and its octave, the meshing frequency range of the fault parts can be concluded through wavelet packet band energy monitoring, during which the wavelet packet decomposition was put forward to identify the fault features of the vibration signals. Considering that the cepstrum can separate and extract the period component of the dense modulating signals which were difficult to identify, and that it also can recognize the speed-frequency of the fault parts, we diagnosed the type and location of the fault parts through meshing and speeding frequency concluded from the utilization of these two kinds of spectrum analysis method. Experimental research has proved that the proposed method can diagnose the gear crack fault accurately, and moreover, this method can be applied to monitor the degraded condition in complex environment of wind turbine gear and thus preventing major faults from occurring such as broken teeth.