AbstractA set of the North American Regional Climate Change Assessment Program (NARCCAP) regional climate models is used in crop modeling systems to assess economically valuable agricultural production in the southeast United States, where weather/climate exerts strong impact on agriculture. The maize/peanut/cotton yield amounts for the period of 1981–2003 are obtained in a regularly gridded (~20 km) southeast U.S. using (a) observed, (b) a reanalysis, and (c) the NARCCAP Phase I multimodel data set. It is shown that the regional‐climate model‐driven crop yield amounts are better simulated than the reanalysis‐driven ones. Multimodel ensemble methods are then adopted to examine their usefulness in improving the simulation of regional crop yield amounts and are compared to each other. The bias‐corrected or weighted composite methods combine the crop yield ensemble members better than the simple composite method. In general, the weighted ensemble crop yield simulations match marginally better with the observed‐weather‐driven yields compared to those of the other ensemble methods.