Aiming at the problems of low classification accuracy caused by the phenomenon that homogeneous pixels have different spectrum in the hyperspectral image and the characteristics of edge pixels being easily confused when combining spatial and spectral information, a method based on hierarchical guidance filtering and nearest regularized subspace is proposed in this paper. Firstly, the principal component of the hyperspectral image is obtained by principal component analysis, and then the hierarchical guidance filtering is performed with the guidance image, the first principal component. The edge-preserving characteristic of the guided filtering, effectively prevents the mixing of spectral information in edge area, and reduces the difference of the homogeneous spectrum at local regions. Finally, the nearest regularized subspace classifier is applied to classify the preprocessed hyperspectral image. Compared with the existing methods on Indian Pines, Salinas and GRSS_DFC_2013 hyperspectral datasets, the results show that the method proposed in this paper has achieved overall classification accuracy of 98.63%, 99.13% and 99.42% on the three datasets respectively, with better classification accuracy and visualization.