Higher temperatures caused by future climate change will bring more frequent heat stress events and pose an increasing risk to global wheat production. Crop models have been widely used to simulate future crop productivity but are rarely tested with observed heat stress experimental datasets. Four wheat models (DSSAT-CERES-Wheat, DSSAT-Nwheat, APSIM-Wheat, and WheatGrow) were evaluated with four years of environment-controlled phytotron experimental datasets with two wheat cultivars under heat stress at anthesis and grain filling stage. Heat stress at anthesis reduced observed grain numbers per unit area and individual grain size, while heat stress during grain filling mainly decreased the size of the individual grains. The observed impact of heat stress on grain filling duration, total aboveground biomass, grain yield and grain protein concentration varied depending on cultivar and accumulated heat stress. For every unit increase of heat degree days (HDD, degree days over 30°C), grain filling duration was reduced by 0.30% to 0.60%, total aboveground biomass was reduced by 0.37% to 0.43%, and grain yield was reduced by 1.0% to 1.6%, but grain protein concentration was increased by 0.50% for cv Yangmai16 and 0.80% for cv Xumai30. The tested crop simulation models could reproduce some of the observed reductions in grain filling duration, final total aboveground biomass, and grain yield, as well as the observed increase in grain protein concentration due to heat stress. Most of the crop models tended to reproduce heat stress impacts better during grain filling than at anthesis. Some of the tested models require improvements in the response to heat stress during grain filling, but all models need improvements in simulating heat stress effects on grain set during anthesis. The observed significant genetic variability in the response of wheat to heat stress needs to be considered through cultivar parameters in future simulation studies. This article is protected by copyright. All rights reserved.