Three forecasting models for Stewart's disease (Pantoea stewartii subsp. stewartii) of corn (Zea mays) were examined for their ability to accurately predict the prevalence of Stewart's disease in Iowa at the county level. The Stevens Model, which is used as a predictor of the early wilt phase of Stewart's disease, the Stevens-Boewe Model, which predicts the late leaf blight phase of Stewart's disease, and the Iowa State Model that is used to predict the prevalence of Stewart's disease, all use mean air temperatures for December, January, and February for a preplant prediction of Stewart's disease risk in a subsequent season. Models were fitted using weighted binary logistic regression with Stewart's disease prevalence data and air temperature data for 1972 to 2003. For each model, the years 1972 to 1999 (n = 786 county-years) were used for model development to obtain parameter coefficients. All three models indicated an increased likelihood for Stewart's disease occurring in growing seasons preceded by warmer winters. Using internal bootstrap validation, the Stevens Model had a maximum error between predicted and calibrated probabilities of 10%, whereas the Stevens-Boewe and Iowa State models had maximum errors of 1% or less. External validation for each model, using air temperature and seed corn inspection data between 2000 and 2003 (n = 154 county-years), indicated that overall accuracy to predict Stewart's disease at the county level was between 62 and 66%. However, both the Stevens and Stevens-Boewe models were overly optimistic in predicting that Stewart's disease would not occur within specific counties, as the sensitivity for these two models was quite low (18 and 43%, respectively). The Iowa State Model was substantially more sensitive (67%). The results of this study suggest that the Iowa State Model has increased predictive ability beyond statewide predictions for estimating the risk of Stewart's disease at the county level in Iowa.