Management of spatial variability of crop yields rely on the availability of affordable and accurate spatial data. Yield maps are direct measure of the crop yields, however, costs and difficulties in collection and processing to generate yield maps results in poor availability of such data in Australia. We used historical mid-season normalised difference vegetation index (NDVI) generated from Landsat imagery over 5 years. Using regression analysis, the NDVI was compared to the actual yield map from a 257 ha paddock. The difference between actual and predicted yield showed that 77% and 93% of the paddock area had an error of <20% and <30%, respectively. The regression function obtained in the paddock was used to simulate crop yield for an adjoining paddock of 162 ha. On an average of 5 years, the difference between actual and simulated yield showed that 87% of the paddock had an error of <20%. However, this error varied from season to season. Paddock area with <20% error increased exponentially with decreasing in-crop rainfall. Furthermore, the error in simulating crop yield also varied with the soil constraints. Paddock zones with high concentrations of subsoil chloride and surface soil exchangeable sodium percentage generally had higher percent of error in simulating crop yields. The simulated yield mapping methodology offers an opportunity to identify within-field spatial variability using satellite imagery as a surrogate measure of biomass. However, the ability to successfully simulate crop yields at farm scale or regional scale requires wider evaluation across different soil types and climatic conditions.