Increasing wheat yield is a key global challenge to producing sufficient food for a growing human population. Wheat grain yield can be boosted by exploiting heterosis, the superior performance of hybrids compared with midparents. Here we present a tailored quantitative genetic framework to study the genetic basis of midparent heterosis in hybrid populations derived from crosses among diverse parents. We applied this framework to an extensive data set assembled for winter wheat. Grain yield was assessed for 1,604 hybrids and their 135 parental elite breeding lines in 11 environments. The hybrids outperformed the midparents by 10% on average, representing approximately 15 years of breeding progress in wheat, thus further substantiating the remarkable potential of hybrid-wheat breeding. Genome-wide prediction and association mapping implemented through the developed quantitative genetic framework showed that dominance effects played a less prominent role than epistatic effects in grain-yield heterosis in wheat.