The traditional sparse representation classification methods only exploit the sparse property while they ignore the neighborhood similarity information in hyperspectral image. To address this problem, a novel sparsity-neighborhood metric classification method was proposed in this paper. Firstly, the proposed algorithm utilizes sparse representation to reveal the sparse properties of data, the following sparse similarity can be calculated in each class of samples. Then, according to neighborhood information, the method constructs the sparsity-neighborhood similarity relationship in each class of samples. Finally, the land cover types can be obtained with the federated sparsity-neighborhood similarity. The proposed algorithm possesses sparse property and neighborhood information, which can enhance the discrimination among different land cover classes to improve the classification performance. The experiments were performed on the Indian Pines and PaviaU hyperspectral data sets. Experimental results demonstrate that the proposed algorithm has better classification accuracy than other algorithms, the overall classification accuracies reach 81.69% and 86.59%, respectively. The proposed algorithm can obtain more homogeneous regions and improve classification accuracy and Kappa coefficient.