The Human Genetics Unit (HGU) of the Medical Research Council (MRC) in Edinburgh has developed the Edinburgh Mouse Atlas, a spatial temporal framework to store and analyze biological data including 3D images that relate to mouse embryo development. The purpose of the system is the analysis and querying of complex spatial patterns in particular the patterns of gene activity during embryo development. The framework holds large 3D grey level images and is implemented in part as an object-oriented database. In this paper we propose a layered architecture, based on the mediator approach, for the design of a transparent, and scalable distributed system which can process objects that can exceed 1GB in size. The system's data are distributed and/or declustered, across a number of image servers and are processed by specialized mediators. © Springer-Verlag Berlin Heidelberg 2005.