For the problem that the monitor signal of rotating machinery fault is complex and hard to extract, a novel diagnosis approach based on Laplacian Eigenmap (LE) for rotating machinery fault is proposed. Extracting and converting the monitoring signal of three typical faults for rotating machinery, and getting 26 time domain and frequency domain features. In the high dimension feature space constructed by those features, using LE algorithm for feature fusion, and extracting fault essence and regularity hidden in the high dimensional feature space to identify incipient fault type. Using two-dimensional or three-dimensional image to showing the low dimensional results extracted, and taking sample recognition rate, between-class scatter and within-class scatter of cluster analysis method as measure indexes, analyzing them from the perspective of pattern recognition. The results show that: Compared with the principal component analysis (PCA) and kernel principal component analysis (KPCA), LE can better extract effective features from high-dimensional feature space to present equipment running status, and realize classification and identification of rotating machinery incipient fault.