One of the most common uses of classified satellite images is mapping land cover alterations via change detection. If images for two different dates are not available from the same sensor, one has to use two sensors with different resolutions, and images or maps must be either upscaled or downscaled. We present a study based on two landcover maps produced for the agglomeration of Geneva, Switzerland: one derived from orthophotos at 0.25m resolution and one from a SPOT image at 5m. Four quantitative methods (“direct”, “nearest neighbour”, “majority” and “statistical”) have been tested to degrade the higher-resolution map to a 5m resolution with a minimum loss of information, in order to make both maps comparable. Upscaled maps have been compared with the original (a) qualitatively by visual examination, and (b) quantitatively by computing Kappa coefficients.Special attention was given to the preservation of thin linear features, such as roads. Visual inspection showed that the majority and statistical methods are unable to preserve thin linear features.With the exception of the direct method,which is the least favourable one, the others perform almost equally well according to the statistical analysis. Overall, the nearest neighbour approach yields the smallest loss of information, with Kappa values up to 92%.