Fast kurtosis specturm has fast detection ability for transient impulse signal and is widely used in fault feature extraction of rotating machinery. A process of extracting mechanical fault feature based on fast computation of kurtogram is proposed and which is applied to the fault feature extraction of composite planetary gears of a tank gearbox. Firstly, the signals of gear normal and spalling fault under specific conditions are analyzed, and compared with the traditional envelope analysis results, it shows that the fault feature extraction method based on fast kurtosis specturm can significantly enhance the amplitude of fault feature frequency. In order to further verify the adaptability of test points and the effectiveness of feature extraction method, 32 test conditions are designed considering gear, rotating speed and load. The test data of each condition are analyzed, and the influence of operating parameters such as rotating speed and load on fault feature extraction is studied. The results show that the fault feature frequencies can be effectively extracted from the data collected by the selected measuring points under various working conditions, and the increase of rotational speed and load is helpful to the fault feature extraction.