In this paper, the seismic waveform was decomposed by the empirical mode decomposition (EMD) method, and the time-domain features of the decomposed intrinsic mode function (IMF) were extracted.The two types of event sources (natural earthquake and artificial explosion) were classified and identified by the extracted features.The results showed that the time-domain features extracted from IMF have good discrimination and recognition ability.The original waveform signal was decomposed into 7.IMFs and residual functions by the EMD, and 2 6.time-domain statistical features were extracted from the original waveform, each IMF, and each residual function, respectively, to form 9.feature groups (named Q0, Q1,…, Q8).Then 7.energy ratio features were calculated from the amplitude-energy ratio of 7.IMFs, and 3 2.features were selected from the time-domain features of the first 4.IMFs, which formed a feature group with 3 9.features (named Q9).A series of identification experiments were conducted on single group or various combinations of the 1 0.feature groups by using the symmetrical Kullback-Leibler (KL) divergence.In each experiment, the training and testing samples were the same, but were randomly selected from the corresponding feature groups of all waveform of some events (30%, 60%, 70, or 90%).The experiments were repeated many times, and the results showed that the time domain features extracted by the second IMF have the best recognition effect, and the correct recognition rate is above 90%.It suggests that the IMF, which has better ability to distinguish event types than the original waveform, can provide more effective features for the recognition of event source type.