The full spatial information cannot be obtained by single filter in characteristics extraction of hyperspectral image. We propose a classification method which combines two kinds of information extracted by non-local means filter and guided filter. This method advances a fusion of spatial information for hyperspectral image classification. One kind of spatial information for all bands of hyperspectral image is extracted by the non-local means algorithm, and another kind of spatial information is obtained by guided filter for the same image after reducing dimensionality with principal component analysis (PCA). Two kinds of spatial information are combined, and the classification is done by support vector machine (SVM). Experimental results show that the proposed algorithm is better than the spectrum information, PCA dimensionality reduction, spatial-spectral SVM, edge-preserving filtering and recursive filtering methods, and the classification accuracy of hyperspectral image is effectively improved.