The rolling bearing is one of the most important and easiest faulty components in rotating machinery. If the performance of rolling bearing can be monitored and evaluated in real-time,the maintenance strategy can be completed in time. The autoregressive（ AR） model is set up and the autoregressive coefficients and residuals of the early failure-free signal and the failure signal with the same type and the same position of bearing failure are extracted. The fuzzy C-means（ FCM） is established by using the early failure-free features and the failure characteristics of the same bearing. The normal and failure cluster center are got. The test data are put into the FCM model by keeping the model invariant and continuously iterating,and the performance degradation index is obtained. Then the empirical model decomposition and Hilbert envelope demodulation are used to verify the conclusions. The experimental results show that the presented performance degradation method is consistent with the results obtained from the accelerated fatigue test.