Agricultural systems are complex and dynamic, being made up of inter-acting bio-physical and human sub-systems. Moreover, agricultural systems are re-markably diverse, both within geographic regions and across regions. Accordingly, this paper focuses on dynamics and heterogeneity in coupled, multi-disciplinary simulation models of agricultural systems. We begin with a discussion of the principal features of agricultural production systems. We then present an example of a ‘loosely coupled’ model, the type of model most researchers have used to represent agricultural systems. We discuss the loosely coupled model's features and limitations, and show how it can be modified to incorporate feedbacks among sub-models. Finally, we use a case study of a hillside production system in Ecuador to illustrate the importance of model coupling, dynamics and heterogeneity in the analysis of production systems. This example shows that feedbacks and threshold effects are most important at sites most vulnerable to tillage erosion.