Waveform signals of earthquake and explosion are nonlinear and non-stationary. A BP (Back-Propagation) neural network model is established to simulate the waveform signal of the earthquake and explosion based upon the real wave data on the Matlab 7.0 experiment platform. It is shown that the differences between the waveform signal simulated by BP neural network and the real waveform signal are quite small. The error percentage of earthquake calculated by the formula 100 %×( predicted value- original value)/original value is less than 1.5% and the error percentage of explosion is less than 1%. The waveform signal simulated by the model can highly correctly reproduce waveform signal's characters of the real earthquake and explosion. This provides a potential way to extract discriminative features of earthquake and explosion by means of network type, network model parameters and network structure. So, it highly deserves further researches.