While crop models often are tested against long‐term mean grain yields, models for aiding decision making must accurately simulate grain yields in extreme climatic conditions. In this study, we evaluated the ability of a general crop model (ALMANAC) and two crop‐specific models (CERES‐Maize and SORKAM) to simulate maize (Zea mays L.) and sorghum [Sorghum bicolor (L.) Moench] grain yields in a dry growing season at several sites in Texas. The root mean square deviation values were 0.36 Mg ha−1 for sorghum with ALMANAC, 0.71 for sorghum with SORKAM, 0.56 for maize with ALMANAC, and 0.83 for maize with CERES‐Maize. For maize, values for coefficient of determination (r2) between measured and simulated grain yields were 0.95 for ALMANAC and 0.88 for CERES‐Maize. For sorghum, r2 values were 0.86 for ALMANAC and 0.45 for SORKAM. ALMANAC and SORKAM should be useful tools to simulate dryland sorghum in drought, as indicated by their root mean square deviation values of <0.8 Mg ha−1. The mean errors for irrigated maize were 2.0% for CERES and 6.2% for ALMANAC. For dryland maize, mean errors were 6.2% for ALMANAC and −2.2% for CERES. In CERES, simulated leaf area index (LAI) and kernel weight appeared to be overly sensitive to drought stress. Further study on the response of LAI and kernel weight to drought in CERES would be valuable. The soil, weather, and crop parameter data sets developed for this study could be useful guidelines for model applications in similar climatic regions and on similar soils.