In this paper we describe a procedure for disaggregating agricultural land use choices at NUTS 2 level to about 18.000 homogeneous spatial units completely covering the usable agricultural area of the EU. The disaggregation procedure uses 40.000 sampling points and aggregate data from administrative regions and requires two steps: First, we employ crop specific, spatial binary choice models to regress land use decisions on local natural conditions (soil, relief, climate) based on the sample information. Results allow predicting crop shares in each spatial unit. Second, consistency with data from administrative regions is achieved by maximising the posterior density of crop shares subject to aggregating equations using the forecast distributions as prior information. Comparison with actual crop shares shows the validity of the procedure.