Aiming at the problem that it is difficult to extract the early weak fault of rolling bearing under the background of noise, a new method combining improved Fourier decomposition (IFDM) and fast spectral kurtosis was proposed to accurately and quickly identify the fault features components. Similar to the intrinsic mode functions generated by empirical mode decomposition, Fourier decomposition method can be used to adaptively decompose the fault signal into a series of intrinsic frequency band functions (FIBFs) which have the physical significance, but its disadvantage is that it cannot control the number of FIBFs. If the signal itself has too many modulation components or the influence of noise is too large, the directly using Fourier decomposition (FDM) will produce a large number of meaningless FIBFs, resulting in a large amount of calculation time waste and increasing the difficulty of data processing. Therefore, an improved FDM method was proposed, in which the fast kurtosis method was used to preprocess the fault signal, and the optimal parameters of the filter and the frequency band of the fault were determined adaptively. Then, only the frequency band with IFDM was decomposed, so the fault frequency components could be accurately extracted and the calculation amount was greatly reduced. The simulation and analysis results of the actual bearing fault signal demonstrate that this method can be used to extract the fault feature frequency more accurately and effectively.