The transmission path of vibration is time-varying， and the picked vibration is often complex because the vibration components are coupled with each other in planetary gearboxes. The early fault of planet bearing related vibration components are weak， and it is often submerged in background noise and strong meshing vibration of gears， which make it difficult to extract the fault feature of planet bearing. Thus， the fault feature extraction scheme of planet bearing is proposed based on cepstral pre-whitening （CPW） and spectral correlation density （SCD） analysis. Firstly， the energy amplitude of the vibration component with strict periodic characteristics is effectively reduced by the CPW. Secondly， the optimal demodulation band parameters are determined by the fast kurtogram algorithm， and the corresponding complex envelope spectrum is extracted. Then， the second-order spectral correlation density analysis is used to extract the planet bearing inner race fault component based on the second-order cyclic characteristics of the bearing. Finally， the features can be extracted by envelope spectrum analysis. It is proved that this method can effectively extract fault features of planet bearings by analyzing the vibration of a planet bearing with a fault on the inner race.