Two hyperspectral image classification algorithms based on Gabor features and locality-preserving dimensionality reduction are proposed. The Gabor transform is studied and implemented to extract features for hyperspectral image in the principal component analysis-projected domain. To protect locality information of neighbor features, locality Fisher discriminant analysis or locality-preserving non-negative matrix factorization is employed to reduce the dimensionality of Gabor-based feature space. The Gaussian mixture model classifier is used for classification results. Experimental results obtained from two hyperspectral datasets show that the proposed algorithms not only extract spectral-spatial features effectively, but also preserve local-feature information and multi-model structure of hyperspectral image. Compared with several existing algorithms, the proposed algorithms can obtain high classification accuracy and Kappa coefficient, and has strong robustness in Gaussian noise environment.