Abstract: In order to reduce the work load of artificial recognition of mine microseismic events, a method for feature extraction was proposed based on the wavelet packet decomposition (WPD) and singular value decomposition (SVD). Firstly, blasting vibration, rock fracture, mechanical interference and electrical interference signals were decomposed into 4 layers based on WPD, then the singular value was obtained by calculating the wavelet packet coefficients on the fourth layer through the use of SVD.With the singular value as the eigenvalue, a 16-dimensional eigenvector was established.The support vector machine (SVM) was adopted to train and classify the microseismic signals in 400 sets of mines. The results showed that compared with blasting vibration, rock fracture and electrical interference, the singular value of mechanical interference signal was the most significant. The classification accuracy of SVM reached 94.5%, which achieved the ideal classification effect.