As far as the vibration signal processing is concemed, composition of vibration signal re-sulting from incipient localized faults in gearbox is too weak to be detected by traditional detectingtechnology available now.The method, which includes two steps: vibraton signal from gearbox is firstprocessed by synchronous average sampling technique and then it is analyzed by complex continuouswavelet transform to diagnose gear fault, is introduced. Two different kinds of faults in the gearbox, i.e.shaft eccentricity and initial crack in tooth fillet, are detected and distinguished from each other suc-cessfully. As far as the vibration signal processing is concemed, composition of vibration signal re-sulting from incipient localized faults in gearbox is too weak to be detected by traditional detectingtechnology available. The method, which includes two steps: vibraton signal from gearbox is firstprocessed by synchronous average sampling technique and then it is analyzed by complex continuous wavelet transform to diagnose gear fault, is introduced. Two different kinds of faults in the gearbox, ieshaft eccentricity and initial crack in tooth fillet, are detected and distinguished from each other suc-cessfully .