Quantitative models that predict cyanotoxin concentrations in lakes and reservoirs from nutrient concentrations would facilitate management of these resources for recreation and as sources of drinking water. Development of these models from field data has been hampered by the high proportion of samples in which cyanotoxin concentrations are below detection limits and by the high variability of cyanotoxin concentrations within individual lakes. Here, we describe a national-scale hierarchical Bayesian model that addresses these issues and that predicts microcystin concentrations from summer mean total nitrogen and total phosphorus concentrations. This model accounts for 69% of the variance in mean microcystin concentrations in lakes and reservoirs of the conterminous United States. Mean microcystin concentrations were more strongly associated with differences in total nitrogen than total phosphorus. A general approach for assessing this and similar types of models for their utility for guiding management decisions is also described.