It is apparent that the surface types and percentage covers recorded in any land use data archive will be a function of both the scale of the source maps and the resolution of the archive itself. This paper illustrates this fundamental cartographic fact in the context of climate modelling with a simple investigation in which the percentage cover of seven basic land use classes was calculated using national survey maps with a scale of 1:50000. The results were compared with similar computations using maps of other scales and with the information contained in two recently published global archives of land surface type. The assessed extent of urban areas is a function of the base map type used. The existence of open water and swamp/marsh areas is not recorded in coarse resolution data archives even when these areas cover 15% of a 1° × 1° grid element. Both these results are features of the data aggregation problem fundamental to geographical representation. This problem cannot be removed simply by producing global data sets at alternative resolutions. A more careful assessment of the sensitivity of models to aspects of the information archive is required.