To extract the weak fault feature of gear, aFractional Fourier Transform(FRFT) filter method based on frequency tracking (FFFT) is proposed to peel the feature components that including fault information by tracking frequency and closely to curve feature of frequency under transient condition. Firstly, the linear multi-scale segmentation method is researched to divide the curve frequency signal into some segments whose frequency are similar to linear; Then, the method to ascertain the filtering parameters by matching the frequency is researched to calculate the FRFT filtering parameters of each segmental signal, and the signal in every segmentation is filtered by FRFT filter using these parameters to complete the FFFT method. The vibration signal of gearbox’s acceleration and deceleration process is analyzed by FFFT and the filtered signal is demodulated, the experimental results show that : linear multi-scale segmentation method is able to divide any curve frequency signal into almost least segments whose frequency are similar to linear; the method to ascertain the filtering parameters by matching the frequency is able to accurately ascertain the FRFT filter parameters of each segmental signals without the influence of the vibration source and the quantity of multi-components; the feature component including fault information of gearbox’s transient condition is accurately extracted by this filter method, the other components and noise are peeled at the same time. The early fault of gear is exactly extracted by the demodulation of extracted feature component, which is difficult to be identified by the traditional method.