Content-based image retrieval relies on low-level image features such as color, texture and segmentation. Humans, however, search for images by their cognitive, deep meaning content. This paper introduces an approach and an algorithm for cognitive image retrieval. Each image is indexed by a visual object-process diagram (VOPD) that represents the image content at the cognitive level. Querying amounts to generating a VOPD that expresses the cognitive content of the sought image. Employing set-theoretic and graph-matching techniques, the algorithm ranks images in the database by their cognitive proximity to the query. A query example illustrates these new concepts.