Abstract: To solve the problem of the low recognition rate of radar signal pulse modulation method in low signal-to-noise ratio (SNR), a method of radar signal recognition based on singular value entropy and fractal dimension is proposed. First, the timefrequency image of the signal is obtained by Choi-Williams distribution, and the singular value entropy of the time-frequency image is extracted. Then the box dimension and the information dimension of the signal spectrum are extracted to form the three-dimension feature vector. Finally, a classifier based on support vector machine is used to realize the classification and recognition of radar signals. The simulation results of eight typical radar signals show that the proposed method is robust to noise and has a high recognition rate. When the SNR is higher than 1dB, the average recognition rate can reach above 95%.