Abstract: In order to recognize the collapsing coal-rock in a fully mechanized caving face,the vibration signals caused by the impact of collapsing coal-rock and hydraulic support tail beam were collected at the scene. A recognition meth- od based on fractal box dimension and Wavelet packet energy moment was proposed. In the method,combining the whole description of nonlinear signal by fractal box dimension with the detailed description at different frequency bands by Wavelet packet moment,the fractal characteristics of vibration signals are analyzed and the box dimensions are cal- culated,then the energy moments at different frequency bands are calculated after the Wavelet packet transform of the vibration signals. The combination of fractal box dimension and Wavelet packet energy moments are used as feature vectors and the input of neural network to identify the two conditions-top coal collapse and roof rock collapse. The ex- perimental results show that the feature vectors can be used to recognize the collapsing coal-rock and the identification rate reaches 95% .