In this research, the authors extract features from intermediate frequency band radar signals in the time–frequency domain for classification. The extracted features are classified via support vector machine and K-nearest neighbour classifiers. They show the accuracy of classification is above 99% for different classes of radar signals except for frequency shift keying signal with accuracy 83% in negative signal-to-noise ratio (SNR). To identify the radars with the same class, the classification accuracy is 91% for SNR between 5 to 15 dB and 64% in the worst case for SNR between −1 to 10 dB. The proposed method is compared with some methods based on the empirical mode decomposition (EMD), cumulant and Zhao Atlas Mark Distribution (ZAMD). The results show that the classification error in the proposed method is less than that of EMD method 55% in the best case and 9% in the worst case. The performance of the cumulant-based method is weaker than that of the proposed method in common designed scenarios becoming almost similar only in one scenario. The ZAMD-based method could only distinguish the signals with different modulations in high SNR while it is unable to classify the signals with the same modulation but different parameters.