Aiming at the high-speed train running state monitor, a high-speed train running state recognition method that coupled wavelet packet energy entropy with fuzzy gray correlation degree techniques is proposed in this paper. Firstly the vibration signals, which are acquired by ten sensors at the key position of high-speed running train, are uniformly segmented and then decomposed by using multi-layer wavelet packets. The wavelet packet energy entropies that are extracted from the vibration signals are used as the fault features. The average energy entropies of 10 pieces of random data of every running state are used as the reference sequences and the entropy energy entropies of other data are used as the detected sequences. By analyzing the fuzzy gray correlation between the reference sequences and the detected sequences, the membership degree of the detected sequences belong to four trains running states is obtained and so the high-speed train running state recognition is realized. Experimental results show that the proposed method can effectively diagnose four high-speed train running states, especially in the case of small samples and inconspicuous fault features, the proposed method is superior to SVM and PNN.