Abstract: In view of problems that γ ray method is not suitable for working face with no or little radioactive elements in roof and radar detection method has little detection range and serious signal attenuation which were used in current coal and gangue interface recognition technologies, the paper proposed a coal and gangue interface recognition method based on Mel frequency cepstrum coefficient and genetic algorithm. The method uses feature difference of acoustic signal produced by dropping process of coal and gangue to recognize coal and gangue. It uses Mel frequency cepstrum coefficient to process denoised acoustic signal of coal and gangue in frequency domain to extract 32 dimensions feature parameters of the acoustic signal, uses genetic algorithm to make optimal process for the parameters to get the best parameter combination, and uses support vector machine and BP neural network to recognize the best parameters. The experiment results showed that the method can recognize falling state of coal and gangue accurately.