Aiming at the problem of how to use band information reasonably in the hyperspectral image classification, this paper proposes a hyperspectral image classification algorithm based on band statistical information weighted K-means clustering. The algorithm considers not only the quantity of information contained in each band and the correlation between bands but also the importance of each band to different clusters. The band weight is determined by the statistics of information and correlation functions. The statistics of information function is defined according to the entropy, standard deviation and mean value of the band image. The correlation function is defined according to the mutual information of adjacent band images. In order to express the importance of each band to different clusters, the band-category weight is introduced, and its entropy information measurement is defined. The above two types of weights are embedded into the K-means objective function. The final classification result can be obtained by minimizing the objective function. Classification experiments are conducted on Salinas and Pavia Centre hyperspectral images using the proposed algorithm, the traditional K-means algorithm, the PCA + K-means algorithm and the subspace band selection + K-means algorithm, respectively. The results demonstrate that the proposed algorithm is higher than the other three algorithms on overall accuracy and Kappa. It shows that the proposed algorithm can effectively improve the performance of hyperspectral image classification.