Well designed graphical user interfaces offer a high potential to increase the productivity of human users. The necessary condition for such a good performance is that the user interface represents the semantics of the underlying application in a clear and comprehensible way. This means, especially, that not only syntactical layout but also semantic consistency conditions between the various interaction objects have to be presented in a graphical user interface appropriately. This is usually termed semantic feedback. Many representation schemes have been proposed to express the properties of graphical man-machine interfaces. However, many of them concentrate on control flow design and do not easily scale up to realistic problems. In this contribution we propose a graph and graph grammar based approach which addresses the problem of semantic consistency of dialogs in graphical user interfaces. The main emphasis here lies on an appropriate specification of the consistency conditions between the various graphical interaction objects of a man - machine interface. We also propose a modularization technique in order to cope with large systems. We show the expressiveness and modularization features of our approach which is based on an analysis of the relationships between the various grafical interaction objects and conclude our paper with remarks on future work.