One important issue faced by wheat producers is temporal and spatial yield variation management at a within-field scale. Vegetation indices derived from remote-sensing platforms, such as Landsat, can provide vital information characterising this variability and allow crop yield indicators development to map productivity. However, the most appropriate vegetation index and crop growth stage for use in yield mapping is often unclear. This study considered vegetation indices and growth stages selection and built and tested models to predict within-field yield variation. We used 48 wheat yield monitor maps to build linear-mixed models for predicting yield that were tested using leave-one-field-out cross-validation. It was found that some of the simplest models were not improved upon (by more complex models) for the prediction of the spatial pattern of the high and low yielding areas (the within-field yield ranking). In addition, predictions of longer-term average yields were generally more accurate than predictions of yield for single years. Therefore, the predictions over multiple years are valuable for revealing consistent spatial patterns in yield. The results demonstrate the potential and limitations of tools based on remote-sensing data that might provide growers with better knowledge of within-field variation to make more informed management decisions.