Radar emitter sorting rate of current methods is not high and they are sensitive to the Signal Noise Ratio (SNR). In this paper, complexity characteristics are applied to sorting unknown complicated radar signal and a high sorting rate is got. The received signal is pretreatment firstly, then the box dimension and sparseness are extracted and they are used as sorting characteristics. Finally, the sorting is completed by KFCM algorithm. By simulation results, the box dimension and sparseness of pretreatment signal sequence are distinguishable and they are not sensitive to SNR, and the lowest sorting rate of different signal is 87% at SNR=5 dB.