Aiming at the fact that the vibration signal of rolling bearing would exactly display non-stationary characteristics and fault features hard to extracted,a fault method of rolling bearing based on local characteristic-scale decomposition（ LCD）and multifractal detrended fluctuation analysis（ MF-DFA） was proposed. Firstly,the vibration signals was decomposed into several intrinsic scale components（ ISC）. Secondly,the intrinsic features hidden in each major ISC were extracted by using MFDFA,among which the generalized Hurst exponents are selected as fault feature. Thirdly,liner local tangent space alignment（ LLTSA） was applied to compress the high-dimension features into low-dimension features which insensitive to fault. Finally,the support vector machine（ SVM） was employed to diagnosis fault. Experiment results of rolling bearing show that the proposed method can classify typical fault of rolling bearing exactly and has certain superiority.