Groundwater modelling is a data-driven decision making process. To support Rio Tinto’s Pilbara iron ore mines, groundwater models are constructed to inform regulatory approvals, environmental assessment and dewatering planning. In many cases, a groundwater model is required long before the commencement of mining. Models built in the early stages of mine development, therefore, can generally be hydrogeologically data poor, specifically in and around the immediate area of a proposed pit. Construction of the groundwater model is reliant on the only information available, the geological drilling designed primarily to inform ore grade. A case study, using an existing, six-year-old numerical model constructed largely on the geological ‘block’ model was revisited. Validation of the model was attempted using four years of post-original model information but subsequently showed a poor match between observed and the original prediction. The model was simplified to include only a fraction of information from the original geological stratification. Parameterisation was also significantly reduced to reflect current hydrogeological conceptual understanding. Simplification of the original complex multilayer model and subsequent re-calibration were constrained to a predictive envelope bound by two conceptual end members that effectively define the upper and lower boundaries of the model uncertainty. It is concluded that groundwater models should not be built to replicate complicated geological geometry due to the fundamental difference between geology and hydrogeology characterisations. Moreover, simple models with less defined aquifer units have advantage over complex models with regarding to quantifying model uncertainty.