a new method of rolling bearings fault diagnosis is proposed based on Laplacian Eigenmaps Manifold learning and improved Variable predictive model based class discriminate(VPMCD), firstly, the vibration signals are decomposed using Local characteristic scale decomposition (LCD),and extracting the feature of each Intrinsic scale component (ISC) to construct the high-dimensional eigenvectors; then using IE algorithm to dig low-dimensional eigenvectors that contain valid information and have intrinsic regularity in high dimensional data; finally, the low-dimensional eigenvectors are inputted Kriging-Variable predictive model based class discriminate(KVPMCD) classifier for pattern recognition. This method makes full use and effective combination of the advantages of LCD in the signal processing, LE in mining feature information, KVPMCD in pattern recognition, and realizes the entire diagnosis from rolling bearings fault feature extraction to fault identification. Experimental results show that the KVPMCD model based on the LE algorithm can effectively identify work status and fault type of rolling bearing.