In order to improve fault diagnosis effect of bearing used in NC machine, a fault feature extraction and diagnosis method of bearing based on LCD multi dispersion entropy was proposed. The vibration signal was decomposed adaptively with local characteristic-scale decomposition（LCD） to obtain the components in different scales of the original signal. Considering the ability of the dispersion entropy in distinguishing the complexity of different signals effectively, the dispersion entropy of intrinsic scale components（ISCs） by LCD was calculated. Thus the complexity metric in different scales of the original signal was gained, which was consequently taken as the feature parameter to describe different bearing states. The feature parameters were then put into SVM for diagnosing the bearing faults. Bearing different fault type and different fault degree diagnosis results show that the proposed method can improve diagnosis effect and has certain superiority when compared with some other methods.