This paper proposes a bag of end members based novel method for content-based hyperspectral image retrieval. The proposed system exploits high resolution spectral signatures of the distinct materials in the hyperspectral images and consists of two steps. In the first step, query and archive hyperspectral images are represented with a feature vector computed by a novel bag of end members based feature extraction method. In the second step, the similarities between the feature vector of the query image and those of the archive images are initially computed by the use of histogram intersection kernel and then the most similar images are retrieved. The proposed system is tested on a hyperspectral image archive created from an EO-1 Hyperion hyperspectral sensor image. The experimental results show that the proposed system increases hyperspectral image retrieval performance.