This paper presents a study on an adaptable image retrieval system used for underwater target identification. Shape and textural features extracted from contrast and range electro-optical imagery data are used to represent each mine-like or non-mine-like sample image. The retrieval system is an adaptable two-layer network where the first layer is structurally adaptable in response to relevance feedback from expert users, while the second layer is adaptable only when a new class is introduced. Each node in the second layer represents one sample image in the training database. Test results on a large electro-optical imagery database are presented, which show the promise of the proposed system as an adaptable image retrieval system.