In order to obtain the ability of identifying radar signals robustly and precisely in the complicated signal circumstance, radar signals are represented by using radar feature vectors consisting of the time frequency image feature, complexity feature and instantaneous quantity feature. The feature vectors are input into a combinational classifier whose main classifier is Support Vector Machine(SVM), and the signal parameters are estimated. Simulation result shows that the method has good performance, even in the rapid changing signal circumstance of low Signal to Noise Ratio(SNR).