Abstract: Band selection is a common approach to reduce data dimensionality of hyperspectral imagery. A new band selection algorithm for hyperspectral imagery based on Kullback-Liebler (K-L) divergence and spectral divisibility was proposed. Firstly by considering every waveband spectral component as the onedimensional vector, the K-L divergence was used to quantify the information amount among them to select the bands combination with maximum information amount and minimum similarity. Then the spectral divisibility distance between two ground objects was calculated to remove redundant bands from the selected bands combination and make the final selected bands combination contain optimal divisibility. Hyperspectral imagery was tested for classification to testify the effectiveness of this band selection algorithm. As shown in the experimental results, the classification accuracy percentage is 92.2% and Kappa coefficient is 0.88, when using spectral angle mapping on the pesudo color synthesis image of 3 selected bands.