The objective of this paper is to compare land use models based on three different proxies for agricultural land rent: farmers’ revenues; land price and shadow price of land derived from a mathematical programming model. We estimate a land use shares model of France at the scale of a homogeneous grid (8 km x 8 km). We consider five land use classes: (1) agriculture, (2) pasture, (3) forest, (4) urban and (5) other uses. We investigate the determinants of the shares of land in alternative uses using economic, physical and demographic explanatory variables. Data on land use is derived from the remote sensing database Corine Land Cover. We model spatial autocorrelation between grid cells and compare the prediction accuracy as well as the estimated elasticities between different model specifications. Our results show that the three rent proxies give similar results in terms of prediction quality of different models. Our results also show that including spatial autocorrelation in land use models improve the quality of prediction (RMSE indicators). One of our econometric land use models is used to simulate the effects of a nitrogen tax as well as to project land use changes in France under two IPCC climate scenarios.