A new method based on empirical mode decomposition (EMD) and kurtosis is used to recognize the coal-rock interface. Vibration signals of the rock and coal are obtained on the hydraulic support tail beam. The signals are first decomposed using EMD to obtain the intrinsic mode functions (IMFs). The IMFs can be used to decompose the signals to distinguish between rock signals and coal signals by kurtosis. In this process, the low frequency interference is reduced and the high-frequency interference is highlighted. The frequency ranges of coal and rock can be analyzed by Hilbert marginal spectrum of the IMFs. The optimal cut-off frequency of the band pass filter can be found according to the frequency domain characteristics of rock. Finally, the vibration signals are processed by the filter. The results show that the coal signal and the interference signal are shielded, and the rock signal mixed in the coal signal can be recognized. Thus, the coal-rock interface is recognized.