Most environmental issues related to agricultural activities arise at a local scale and need to account for the spatial dimension of processes. We present a multiregional computable general equilibrium model which, by dropping out the classical Armington assumption on trade while explicitly introducing transport costs, is in contrast with the present modelling tradition. It is implemented as a mixed complementarity program using the GAMS software. Simulations are run with an 8-region version of the model, but successful experiments have been carried out with up to 35 regions. We also adopt an original two-step calibration method which permits the endogenous calculation of the multiregional benchmark equilibrium. An unconditional systematic sensitivity analysis regarding elasticity parameters concludes to a satisfactory robustness of the model. Finally, two stylised agricultural policy alternatives with an underlying environmental objective demonstrate the relevance of the multiregional framework for the comparison of scenarios which are defined at different local scales.