Texture is an important property of many types of images, and has been widely used in a number of applications to describe, classify and index images. In this paper, we propose a technique for characterising texture by modelling the relationships between the high and low frequency bands of its wavelet decomposition, which have been experimentally shown to provide information which is not available when considering such bands independently. A similarity measure using the recently proposed earth mover's distance is formulated, and the results of using this measure for retrieval of textures from a database are presented.