Abrupt spatial changes in species composition of Swiss forests reflect small-scale patterns in topographical and geological site factors as well as management influences. Therefore, it has always seemed appropriate to define discrete forest units, i.e. plant associations or vegetation units generated by clustering, rather than by the arrangement of plots along underlying environmental gradients. Well-defined forest units are essential for mapping projects aimed at management applications. After having revised the old classification system of Ellenberg & Klötzli (1972) on the basis of new data, we faced a lively resistance to the new classes, primarily by people who had classified forests in specific regions. In search of a solution to the problem, we tested the old classification system by using direct and indirect gradient analyses on more than 6000 relevés. We found that these forest relevés are arranged continuously. We compared the resulting gradients with secondary site factors. Previously described vegetation units were reviewed by displaying them in ordination space.