Abstract
Synopsis Multidimensional scaling, in the form of principal coordinates analysis and two-way correspondence analysis, is used to illustrate inter-practice variation in patterns of psychiatric diagnoses provided by data from the longitudinal file of the Second National Morbidity Survey. The results strongly support the view that general practitioners' diagnostic habits should be validated before their records are used to provide data on ‘official’ estimates of psychiatric morbidity. It is recommended that, whatever the quality of the data, large tables of official socio-economic or medical statistics should be supplemented by graphical summaries, as they quite often are in France.