Soils play a central role in many environmental issues. If processes are generally studied at local scale by field studies, information at regional levels is needed, especially for decision making, and requires an increasing use of modelling. The models have generally been developed for estimations at local scales. Their use at regional levels is faced with several difficulties: i) a use in contexts that are beyond those used for the model design and validation, ii) the lack of precise input data or parameter calibration data. Soil input data are generally estimated from available spatialized soil data bases. Based on15-years experiences in using the Soil Geographical Data Base of France (SGDBF) at 1:1,000,000 to spatialize the crop model Stics, the thesis objective is to develop a methodology for evaluating spatialized sets of soil input parameters for this model based on a sensitivity analysis of Stics for soil input parameters and an uncertainty analysis for the most sensitive parameters.