In the paper, a new method based on the combination of Hilbert with frequency domain kurtosis (FDK) which can filter out the fault signals and extract the characteristic signal of interest. The Hilbert and FDK will be introduced, and we will optimize the FDK which is more accurate and superior than traditional algorithm in the extraction of fault signals. By applying the method to the practical roller bearing vibration signals, the results show that the method is more advantageous to extract fault signals and reduce the influence of interfering components in the background noise.