The Gibbs distribution is often used to model micro-textures. This includes a definition of a neighbourhood system. If a micro-texture contains a large-scale variation, the neighbourhood system will be large, which implies many parameters in the corresponding Gibbs distribution. The estimation of the parameters for such models will be difficult and time consuming. I suggest, in this paper, a separation of the micro-texture into a large-scale variation and a small-scale variation and model each source of variation with a Gibbs distribution. This method is applied on full-tone print of newsprint to model the variation caused by print mottle. In this application, the large-scale variation is mainly caused by fibre flocculation and clustering and the small-scale variation contains the variation of the fibres and fines on and between the clusters. The separate description of these two variations makes it possible to relate different kinds of paper qualities to the appropriate source of variation.
