The evaluation of cropping system at farm and regional scales requires information and data on potential crop yields and their variation in time and space. Cropping systems models to represent and explore future scenarios need to be able to capture the main limiting factors affecting productivity and the major sources of crop yield variability. For irrigated rice, the main crop grown in the region of Camargue, south of France, these factors are: low temperature and solar radiation during the crop cycle, success of sowing (stand density), nitrogen application, and weed management. As it is often the case, few data are available for analysing and establishing empirical relationship between observed yield and the different limiting factors listed above. The relationships thus established, on the other hand, do not allow exploring alternative systems, not yet practiced, or outside of the range of conditions for which the models were developed. Efforts have been devoted for decades to building comprehensive crop growth models that predict yields given a combination of environmental and management conditions. However, these models do not consider the whole range of limiting factors that determine crop variability in the field. While simulation modelling can be used to estimate a range of variability in crop yields that is deterministic (i.e., due to variation imposed in model parameters), observed crop variability has always a stochastic component, and variable degrees of uncertainty on the sources of such variability. Here, we compare the magnitude and nature of these two types of variability, observed and model-generated, to evaluate to what extent crop simulation models can be used to approximate such reality.