Wavelet deconvolution is one of the main methods to improve the resolution of seismic data,and the key is the accuracy of the wavelet.Conventional methods usually presume that the wavelet is the minimum phase,but this may not be correct.To solve this problem,the authors utilized the property of the higher order statistic quantity that it includes phase information of the system and suppresses Gaussian noise to study the wavelet extraction method based on cepstrum and, as a result,gave the principles of extracted wavelet based on complex cepstrum,bispectrum and trispectra.In addition the authors constructed a smooth window function to improve the accuracy of wavelet.The method does not need to make any assumption about wavelet's phase,and it can extract wavelet of any phase.Tackling the mixed wavelet,the authors made theoretical experiments to verify the effectiveness of the method,and the results of actual data processing showed the effectiveness of the method.