This paper introduces a novel content-based image retrieval (CBIR) system for hyperspectral image databases using both spectral and spatial features computed following an unsupervised unmixing process which minimizes human intervention. The set of endmembers obtained from the image by an Endmember Induction Algorithm provides the image spectral features. Spatial features are computed as abundance image statistics. Both kinds of information are functionally combined into a dissimilarity measure between two hyperspectral images. This dissimilarity measure guides the search for answers to database queries. The system allows the user to retrieve hyperspectral images containing materials similar to the query image, and in a similar proportion. We provide validation results using both synthetic hyperspectral datasets and real hyperspectral data.