Due to the insufficiencies of traditional sampling and quantitative characterization methods for oil shale formation, this paper proposed a new approach for rock classification and sample selection based on the study of the shale oil well section in the Mahu Depression. The approach combines dual-energy CT and MaipSCAN mineral scanning electron microscopy data and takes into account rock density, photoelectric index, and mineral identification. Clustering analysis is utilized for the classification. The proposed method is then applied to identify prevalent formations. An innovative representative rock sample selection method and process are introduced that considers macroscopic heterogeneity. The results show good consistency between the photoelectric index-density classification and minerals, and the clustering classification method is capable of further fine sample selection. Layers with high plagioclase content exhibit better oil and physical properties and can serve as geological and engineering sweet spots. The proposed classification method and sample selection process are reliable and provide new ideas and methods for the systematic evaluation of oil shale reservoirs.