Gas discharge will produce rich electromagnetic, optical as well as acoustic signals. Compared with the other signals, acoustic signals are also significant and would offer non-contact, low cost and easy-operation approach for online discharging monitoring, which require more attention and intensive study. In this paper, we studied the characteristics of acoustic signals in the corona, transient glow, spark, and glow discharging modes generated in a DC pin-to-pin configuration and developed a method using acoustic signals to classify the different discharge modes. The acoustic signals of the discharge at different gaps were recorded by adjusting the gap distance. 250 sets of acoustic signal samples were collected for each discharging mode. It was found that acoustic signals behave differently in different modes. Based on the short-time Fourier transform (STFT) of the acoustic signals, a novel method for discharge mode classification using the support vector machine (SVM) approach was developed. The final predictive accuracy of the trained classifier exceeds 90%.