Abstract: In allusion to the problem that during the analysis on power quality disturbances the frequency band division of empirical Fourier decomposition (abbr. EFD) is not adaptive, an improved FED method was proposed. Firstly, the maximum envelope of normalized spectrum of disturbance signal was extracted by piecewise cubic Hermite interpolation. Secondly, the maximum value of the envelope was searched and it’s dynamic measurement was calculated. Thirdly, the frequency, whose dynamic measurement was greater than the set threshold, was taken as the characteristic frequency of disturbance signal, and the midpoint of adjacent characteristic frequency was taken as the boundary of frequency band segmentation of disturbance signal. Then, the inverse fast Fourier transform (abbr. IFFT) was performed on the divided frequency band to obtain the analytic Fourier intrinsic band function (abbr. AFIBF) of the corresponding frequency band, and then, Hilbert transform (abbr. HT) was applied to the decomposed AFIBF components to extract their disturbance parameters. Finally, through the analysis of simulated signals and real measured data, the correctness and effectiveness of the proposed method were verified.