A new approach for condition evaluation of bearing-rotor based on Kernel Principal Component Analysis and Gaussian Mixture Model is proposed. The wavepacket energy spectrum of signal is obtained firstly. Then, the principal component in feature space is extracted by Kernel Principal Component Analysis. The processed data are used to model the Gaussian Mixture Model and the parameters are estimated by EM algorithm. The equipment condition is evaluated by overlap of Gaussian Mixture Models. The approach is validated by emulational bearing rotor’s vibration data. Testing results show that Kernel Principal Component Analysis can concentrate the energy of the signal in feature space more efficiently. The performance of the overlap of Gaussian Mixture Model based on the proposed method can evaluate bearing-rotor condition more suitable.