Effects of increasing carbon dioxide concentration [CO2] on wheat vary depending on water supply and climatic conditions, which are difficult to estimate. Crop simulation models are often used to predict the impact of global atmospheric changes on food production. However, models have rarely been tested for effects on crops of [CO2] and drought for different climatic conditions due to limited data available from field experiments. Simulations of the effects of elevated [CO2] and drought on spring wheat (Triticum aestivum L.) from three crop simulation models (LINTULCC2, AFRCWHEAT2, Sirius), which differ in structure and mechanistic detail, were compared with observations. These were from 2 years of free-air carbon dioxide enrichment (FACE) experiments in Maricopa, Arizona and 2 years of standardised (in crop management and soil conditions) open-top chamber (OTC) experiments in Braunschweig and Giessen, Germany. In a simulation exercise, models were used to assess the possible impact of increased [CO2] on wheat yields measured between 1987 and 1999 at one farm site in the drought prone region of Andalucia, south Spain. The models simulated well final biomass (BM), grain yield (GY), cumulative evapotranspiration (ET) and water use efficiency (WUE) of wheat grown in the FACE experiments but simulations were unsatisfactory for OTC experiments. Radiation use efficiency (RUE) and yield responses to [CO2] and drought were on average higher in OTC than in FACE experiments. However, there was large variation among OTC experiments. Plant growth in OTCs was probably modified by several factors related to plot size, the use (or not use) of border plants, airflow pattern, modification of radiation balance and/or restriction of rooting volume that were not included in the models. Variation in farm yields in south Spain was partly explained by the models, but sources of unexplained yield variation could not be identified and were most likely related to effects of pests and diseases that were not included in the models. Simulated GY in south Spain increased in the range between 30 and 65 ue to doubling [CO2]. The simulated increase was larger when a [CO2]xdrought interaction was assumed (LINTULCC2, AFRCWHEAT2) than when it was not (Sirius). It was concluded that crop simulation models are able to reproduce wheat growth and yield for different [CO2] and drought treatments in a field environment. However, there is still uncertainty about the combined effects of [CO2] and drought including the timing of drought stress and about relationships that determine yield variation at farm and larger scales that require further investigation including model testing.