Feature extraction and selection are critical for pattern recognition system; they are related with the final identification effect of the system. In the paper, a method of entropy analysis was proposed for extracting the features of signals, which was an important step in the modulation recognition of communication signals. After transforming the signals through FFT and Chirp-Z transform, the Shannon entropy and the index entropy were found as the identification features. The simulation results show that after FFT and Chirp-Z transform, when the SNR is greater than 20dB, 10dB, the entropy value curves are stabilized respectively. Using the entropy analysis method to extract the signal characteristics is less affected by noise, and the entropy curves are relatively stable. It’s more conducive to distinguish the signal modulation and has a very good application value.