Abstract: To deal with the problems of emitter sorting caused by parameter complexity and agility of multi-function radars. The bispectrum images of radar signals are divided equally into some identical regions and singular value decomposition is used,bispectradistribution entropy and singular spectrum entropy are extracted for the cascade features. This method takes advantage of the superiority of bispectrum analysis in restraining the affection of coloured Gauss noise and retaining the amplitude and phase informations.It organically integrates the advantages of the theory of the bispectrum analysis,singular value decomposition theory and informationentropy theory,and fully reflects the intrinsic signal informations. The radar signals are classified by fuzzy C-means for experimentation to demonstrate the validity of this algorithm. Experiments conducted on six typical emitter signals show that the algorithm isof preferably sorting precision under moderate SNR, and surmounts the disadvantages of higher feature dimension and deficientcompactness of clusters about traditional methods for feature extraction of pictures.