Spatial variability in grain yield across a paddock often indicates spatial variation in soil properties, especially in regions like the Victorian Mallee. We combined 2 years of field data and 119 years of simulation experiments (APSIM-Wheat and APSIM-Barley crop models) to simulate crop yield at various levels of N application in 4 different management zones to explore the robustness of the zones previously determined for an experimental site at Birchip. The crop models explained 96% and 67% of the observed variability in wheat and barley grain yields, with a root mean square error (RMSE) of 310 kg/ha and 230 kg/ha, respectively. The model produced consistent responses to the observed data from the field experiment in 2004 and 2005 where a high and stable yielding zone produced the highest dry matter as well as grain yield, while a low and variable zone recorded the lowest grain yield. However, from the long-term (119 years) simulation, the highest median wheat yield value was obtained on the low variable zone (2911 kg/ha) with high N fertiliser application, while the lowest was obtained on the high variable zone (851 kg/ha). Similarly, the highest barley yields (1880–3350 kg/ha) occurred on the low variable zone using the long-term simulation. In 10–20% of years the highest yield occurred in the high-yielding zones, with the variable and stable zones changing rank with interactive behaviour only under early-sown conditions. Our analyses highlight the problem of using a limited range of seasons of different weather conditions in agronomy to make strategic conclusions as the long-term simulation did not confirm the original yield zone determination. The challenge ahead is to predict in advance the seasons where application of N fertiliser will be beneficial.