The recognition of coal-rock interface in the top caving was investigated via the vibration signals of the tail beam of the hydraulic support. Wavelet packet transform was used to process the vibration signals. A newly feature based on wavelet packet energy spectrum was proposed to identify the coal-rock interface in top coal caving. The interface was determined by comparing the features in the two cases of coal dropping and rock dropping. The experimental results show that the energy proportion of vibration signal in high frequency bands when rock dropping is greater than that when coal dropping.