Groundwater models are often overly simplistic representations of complex systems. Predictive parameter uncertainty of simple groundwater flow models can be quantified using an array of non-linear techniques. A comparison of formal Bayesian approaches including GLUE (Generalised Likelihood Uncertainty Estimation), Markov Chain Monte Carlo using DREAM (DiffeRential Evolution Adaptive Metropolis), and Null-Space Monte Carlo is made. A large scale numerical model was developed to simulate groundwater impacts from iron ore mining in Western Australia and was assessed using three predictive uncertainty techniques. Our results demonstrate that GLUE and DREAM produce broader groundwater impacts when compared to Null-Space Monte Carlo results. GLUE and DREAM analysis reveals multi-normal and non-normal parameter distributions using information from the observation dataset. The key is to improve the understanding of the methodology, and to analyse posterior parameter distributions to ensure worse case impacts are fully explored. This study provides a framework to quantify uncertainty using regional models when highly parameterised inversion is not possible.