Groundwater in alluvial terrain, supporting much of the global requirement for irrigation and domestic water, is at risk of sustained water level decline and/or contamination over large areas. In southern Bangladesh excessive arsenic (As) in shallow groundwater has led to deeper groundwater becoming the preferred alternative source of potable water. The vulnerability of deeper tube-wells to As breakthrough from shallow levels can be assessed using groundwater modelling, but representation of alluvial aquifer heterogeneities in large-scale groundwater models presents a challenge; what level of complexity is required? To assess the optimum level of complexity necessary in models of groundwater flow and solute transport in the Bengal Aquifer System (BAS), we explore a range of representations of the lithological heterogeneity using a multi-modelling approach. We use an array of geological information including drillers’ logs (n = 589) and hydrocarbon exploration data (n = 11) across an area of 5000 km2 as a basis for alternative representations of upscaled aquifer heterogeneity, characterising hydrogeological structure in a series of five groundwater models at increasing levels of complexity. We rank the models by comparing model outcomes of travel time with available data on groundwater age based on 14C. The results demonstrate the importance of spatial heterogeneity and suggest the significance of incorporating vertical heterogeneity in model representations of the Bengal Aquifer System and similar spatially extensive fluvio-deltaic aquifers.