Abstract: In order to reduce the degree of cross-contamination between the two components of distortion product otoacoustie emissions (DPOAE) during separation, a time-domain separation algorithm model based on inverse fast fourier transform (IFFT) combined with time window was built and separation error was analyzed, based on which a window duration selection method was proposed for the separation algorithm. According to the time-domain waveform after IFFT, the centroids of the waveform under the two main peaks of the signal envelope were caculated. Then selection of a suitable separation boundary between the two components was identified according to the movement speed of the summed centroids to separate them. Simulation of the separation algorithm combined with the window duration selection method was done and DPOAE data from experiment was seperated. Simulation results and actual DPOAE seperation show that the separation algorithm combined with this method can also separate the two components successfully and greatly reduces the amplitude and phase disturbance caused by cross-contamination between the two components. Therefore the component separation algorithm combined with the window duration selection method is effective and it improves the separation accuracy.