An algorithmfor image and sub-image retrieval is presented which is based on colour pairs occuring in inhomogeneous patches in the query image. The algorithm is a modification of the Multi-modal Neighbourhood Signature (MNS) method developed at the University of Surrey. By quantising into cells the space used to represent the colour pairs and then indexing the cells in a database, significantly faster retrieval times are achieved. Retrieval performance is enhanced and this is also demonstrated using recall-precision graphs for a test database of images.