Thermal heat sum indices are used to predict the duration of corn (Zea mays L.) growth intervals. Surprisingly, there exists very few studies in agronomic literature that assess predictive accuracy of these indices. This paper contributes to this topic in several ways using USDA-NASS corn growth data for Iowa (IA) and Illinois (IL) which together cover about 30% of corn production in the United States. First, the dataset spans from 1996 to 2016 and covers all the districts in both Iowa and Illinois. Second, the data are used to evaluate the predictive accuracy of several types of thermal heat sum indices (bilinear and nonlinear) for granular growth intervals. Third, the evaluation is executed using out-of-sample contemporaneous forecasts. The results show that nonlinear thermal heat sum indices (HSIs) have superior predictive accuracy to bilinear indices, whereas linear HSIs turn out to be inferior. Overall, the predicted durations using nonlinear thermal HSIs for the Iowa and Illinois combined districts are reasonably accurate. There is little bias across most of the growth intervals despite including the 2009 and 2012 growing seasons with extreme weather conditions. The mean absolute percentage error (MAPE) of the predicted durations relative to actual is well within 10% for almost all growth intervals across combined districts in IA and IL with a low MAPE of 5% for the full growing season duration using the nonlinear thermal heat sum indices. In comparison with other growth intervals, the vegetative growth interval has the lowest MAPE for both IA and IL combined districts.