Today, images are used in almost every career, and a different range of professional groups use images in their work. They have always been used as a means to convey information. Images are very often used either alone or in compound with text, as they are able to conveying information difficult to depict in words. However, image retrieval based on syntactic features alone does not provide support for various types of queries. There is a mismatch between what the current solutions for this are able to providing, and what we demand from an image retrieval system. This gap between what is possible and what we require, by many authors has been named the "Semantic gap". The focus of this paper was how to obtain successful semantic image retrieval from large image collections. The research investigated some of the difficulties of getting correlate results when searching in image collections. A new tool, the shape thesaurus, was suggested as a step into improving image retrieval. The principal motivation behind this suggestion has been to assess the possibilities of narrowing the Semantic gap.