Due to the high data dimensionality of a hyperspectral image, dimensionality reduction algorithm has attracted much attention in hyperspectral image analysis. Band selection algorithm, which selects appropriate bands from the original set of spectral bands, can preserve original information from the data and is useful for image classification and recognition. In the present paper, a novel band selection algorithm based on orthogonal projection divergence (OPD) is proposed, it aims to discriminate the interesting objects from background and noise information, maximize the spectral similarity between different spectral vectors by projecting the original data to feature space. Two HYDICE Washington DC Mall images and an HYMAP Purdue campus image data were experimented, and support vector machine (SVM) classifier was used for classification. The selected band number varies from 5 to 40 in order to study the impacts of different band selection algorithms on different features. For the computation complex, the sequential floating forward search (SFFS) was used to get the appropriate bands. The experiments have proved that our proposed OPD algorithm can outperform other traditional band selection methods such as SAM, ED, SID, and LCMV-BCC for hyperspectral image analysis. It is proven that OPD band selection is effective and robust in hyperspectral remote sensing dimensionality reduction.