Large amounts of waste material with high concentrations of various metal(loid)s are common features at old glassworks sites in southeastern Sweden. As epidemiological data indicates increased health risks for local residents around these sites, questions about how exposure occurs have been raised. One important step in assessing site-specific total exposure is to quantify exposure via intake of drinking water, and the critical variable to determine here is the metal concentration in the water to be consumed. In many routine risk assessments this is, despite well-known uncertainties, done by applying conservative and simplified transport models that simulate concentrations in water abstraction wells at certain distances from the source of contamination. To assess the accuracy of such an approch, we used a probabilistic method to simulate metal concentrations in drinking water within 500 m distance from 10 heavily contaminated glassworks sites, by applying the transport equations of the Swedish EPA generic risk assessment model. Exposure was then calculated. When concentrations in drinking water were calculated using this modelling approach, the resulting exposure estimates indicated that as much as 100% of the local residents with private wells may reach a daily intake above applicable toxicological reference values. However, when assessing the risk from measured concentrations in household drinking water instead, not even the 95th percentile of the population reached the toxicological limit. It hence appears as if metals are efficiently sorbed to the solid matrix of the soil at these sites, leaving the drinking water with concentrations well below drinking water criteria. For the average risk assessor working with contaminated sites, who often lacks training in hydrogeochemistry, the contradictory results highlight the need of more realistic generic models to be used in routine assessments of water contamination and exposure via consumption of drinking water.