Radar signal sorting is picking-up pulse serial of same radar emitter from dense complex pulse signal flow. The tolerance of radar signal sorting is analyzed in modern electronic warfare. The complex and dense pulses environment makes it become a vital factor to restrict the efficiency of sorting of the conventional multi-parameters signal sorting system. A segment clustering radar signal sorting method is presented based on support vector clustering (SVC) according to the idea of statistics learning theory. It prevents tolerance from affecting radar sorting. The accuracy of sorting and the sensitivity of algorithm on parameter variation is analyzed. The experimental results show that the sorting method presented is effective to overcome the tolerance of radar signal sorting.