To assess the equipment performance degradation accurately, wavelet packet entropy (WPE) and Gaussian mixture model (GMM) were combined to evaluate the degradation state of bearing performance in this research. Firstly, WPEs, as the feature vectors, were extracted from the bearing vibration signal to describe the bearing running state. Secondly, the feature vectors under normal state were used to construct the Gaussian mixture model as the baseline. Then the GMM of every running state is established during the experiments, and the corresponding deviation value (DV) from the baseline model is calculated to evaluate whether degradation is occurred and degradation level of the bearing. Experiment results indicated that, comparing with the logistic regression based degradation evaluation method, the proposed scheme can depict the bearing degradation process accurately in the full life cycle without the need of historical data or predefining the prior probability of each possible degradation state.