The wavelet signal reconstruction and Hilbert-Huang transformation were used in the fault diagnosis of the reducer. According to the non-Linear and non-stationary characteristics of ZHLR-170 reducer's vibration signal, it found the frequency of the fault signal's maximal ampLitude was about 270Hz with a trait of continuous distribution in time, which could indicate that the reducer got transmission failure at its first-stage and most gears of its input terminal were wear-out. The results show that by the auto-adapted signal processing method, the fault vibration signal's occurred time and occurred cycle could be obtained effectively, also the fault's frequency, time and value distribution, thus could be more effective to recognize the fault type of the reducer.