A quantitative measure for the spatial representativeness of ground level ozone concentrations is described and discussed as a means to compare modeled and measured data. It is derived from the analysis of the routinely measured diurnal variations of more than 300 German sites using principal component analysis. From the number of principal components (six) needed to explain the representative part of the time series (95%, up to the estimated error of the measurements) and the smallest correlation distance of these principal components we infer a radius of representativeness of about 4 km. This is a limit for the resolution of Eulerian models in order to simulate the representative part of the observations. ¿ 1998 American Geophysical Union