Cereal varieties were selected in full sun conditions, and models of crop productivity were also designed and validated for full sun conditions. Both may therefore not be adapted to agroforestry cultivation and simulation, respectively. Durum wheat productivity was monitored in agroforestry systems with deciduous broadleaved tree species during the last ten years in southern France. Dedicated experiments with artificial shade were also set up to control light availability. Yield components (plant density, number of tillers, number of ears, grains per ear, grain weight, protein content of the grains and qualitative parameters) were measured in both full light and reduced light conditions. Beyond a threshold of 70% radiation during the flowering and grain filling stages, the cereal yield was not decreased by the shade. For heavier shade conditions, the cereal yield was significantly reduced. A crop model (STICS) was used to simulate the crop productivity in both full light and shaded conditions. Leaf area indexes were not much reduced by deciduous tree shade, as most the leaf growth was finished before significant shade occurred on the plot. The main impact of the shade was on the fertility of the ears. The grain weight was not modified, while the protein content was significantly increased in shaded conditions. The protein yield per hectare was therefore not much reduced by the shade, even when the dry matter grain yield was significantly reduced. Some processes were not correctly simulated in the shade such as crop temperature and soil temperature, and improvements of the crop model are suggested. The optimization of temperate agroforestry systems would require the selection of cereal varieties less sensitive to a deficit of radiation at flowering, and the genetic variability of cereals should be explored for this new criterion