In order to improve the accuracy of feature extraction, a Multi-scale Fuzzy Entropy algorithm is proposed based on Entropy theory. The new algorithm combines the concept of Fuzzy Entropy and Multi-scale Entropy. Aimed at the shortage of Empirical mode decomposition (EMD), Index Energy is selected as evaluation index, and a new feature extraction method is designed to dispose vibration signal. Experiment shows that the new method can effectively distinguish different condition of rotor vibration signal and can significantly improve the effect of feature extraction.