Assessments of the health effects of pollutants generally use models in which exposure variables and model parameters are point values, often chosen as conservative estimates. A more realistic approach is to characterize the uncertainty of each variable and parameter explicitly as a probability distribution. This paper presents a probabilistic version of an established model for metabolism of ingested uranium and characterizes the parameters of the model as distributions. It then characterizes the uncertainty of the threshold level for nephrotoxic risk of uranium, and uses distributions of the heterogeneity in drinking water intake of uranium in the U.S. to assess nephrotoxic risk. It further evaluates the implications of inadequate data on the heterogeneity of dietary intake.