This paper illustrates the hydrological limitations and underlying assumptions of 4 catchment modelling approaches representing different generic classes of predictive models. These models are commonly used to estimate the impacts of land use and management change on stream flow and salinity regimes within a target region. Three approaches are based on a simple conceptual framework that assumes a single layer groundwater aquifer and requires minimal information and calibration (Zhang-BC2C, CAT1D-BC2C and LUCICAT), whereas the fourth approach (CAT3D) adopts a fully distributed highly parameterised catchment model capable of simulating complex multi-layered groundwater aquifer systems. All models were applied to the Gardiner subcatchment within the Goulburn–Broken region of Victoria, identified as a National Action Plan for Salinity priority subcatchment. Current condition simulation results were compared with observed stream flow and groundwater hydrograph data. Results show that the simple frameworks predicted whole-of-catchment mean annual salt and water yield with minimum parameterisation. The fully distributed framework produced similar catchment-scale responses to the simple approaches, but required more intensive input data and solution times. However, the fully distributed framework provides finer temporal and spatial scale information within the catchment. The more detailed models (such as CAT3D) also have the predictive capacity to assess the within-catchment dynamics at a range of scales and account for landscape position and complex surface/groundwater interactions.