The development of multi-agent models for agriculture has allowed the inclusion of farm decision-making behaviour and interactions in the simulation of smaller agricultural regions. Important methodological impact for this has come in particular from scientists from Germany. Currently under construction, the SWISSland model claims to depict as realistically as possible the 50,000 family farms comprising the whole of Swiss agriculture in all their heterogeneity as regards farm and cost structures as well as farm decision behaviours, with the aim of improving the simulation and forecasting of structural change. This paper describes methodological aspects in the formation of the agent population by combining various data sources such as accounting and spatial data and the results of surveys. As its basis, SWISSland uses the 3300 Swiss Farm Accountancy Data Network (FADN) farms, whose representativeness is substantially improved by means of a corrective procedure. Individual-farm optimisation models simulate the heterogeneous behaviour of the agents, for whom a potential exists for land trade within regional groups. With the linking of different methods and recorded data, we can expect to see a marked increase in the quality of the assessment of policy consequences.