Abstract: To accurately locate and classify the power quality disturbance, a novel improved Hilbert-Huang transform (abbr. HHT) based power quality disturbance method was proposed. The amplitude-frequency parameters obtained by traditional HHT were processed by the extreme value sliding window averaging algorithm to conduct the ridding the extremum value and equalization, so the accuracy of judging the time of occurrence and the time of end of the disturbances in Hilbert-Huang spectrum was improved. By means of the new method the amplitude-frequency curves, the Hilbert-Huang spectrum and Hilbert marginal spectrum were determined and from them four characteristic quantities of the disturbance, i.e., frequency components, durations, voltage amplitudes and the amplitude of Hilbert-Huang spectrum were extracted to implement the classification and identification of the disturbance. Simulation results show that the combination of improved HHT with decision-making tree not only suit to the location and classification of single disturbance, but also it can obtain good results for complex non-stationary disturbances, thus it possesses a certain noise-proof capability.