Crop modelling is a critical part of integrated regional assessments in AgMIP. It needs to be undertaken with multiple crop models to characterise uncertainty in predictions of production. To ensure consistency, it also needs to follow consistent protocols that specify the methods for calibrating varieties, and application of models to simulate crop yields in farms under current climate and management, and under a range of climate change and adaptation scenarios. High quality data are needed from experiments to calibrate the models. These experiments are called sentinel sites, and should include measurements of growth stages and yield, as well as information on soils, climate and management. The calibrated models are then used to simulate crop production in farmers’ fields to provide the primary yield input to the economic modelling. Information is needed on soils, management and yields currently achieved in these fields. Three types of simulations include yields under current climate and management, yields under future climates with current management, and yields under future climate with adaptation (and/or greenhouse gas mitigation) strategies. Protocols have been defined to guide each of these steps, from what defines a sentinel site, to the information needed to interpret simulated yields and provide necessary data for economic models. The application of multiple crop models is supported by the AgMIP IT tools, which port shared information between different models and assist in the generation of management detail for the farmers’ fields, which can be too numerous (e.g. >100) to be practically done by hand. The successful application of the crop modelling protocols and supporting tools is clearly demonstrated by the recent results from the AgMIP regional studies. However, a clear challenge for ApMIP is to move from analysing cropping systems to agricultural systems, which can include the integration of crops and livestock on farms.