1. Ecological theories suggest that higher plant genetic diversity can increase productivity in natural ecosystems. So far, varietal mixtures, that is, the cultivation of different genotypes within a field, have shown contrasting results, notably for grain yield where both positive and negative mixing effects have been reported. Such discrepancy between ecological theories and agronomical applications calls for a better understanding of plant–plant interactions in crops. 2. Using durum wheat Triticum turgidum ssp. durum as a model species, we investigated the effect of functional trait composition on productivity and grain quality of varietal mixtures by growing 179 highly diverse genotypes in pure stands and 197 two‐way mixtures in field conditions. We quantified the agronomic performance of the mixtures relative to their components grown in pure stands on two variables related to productivity, vegetative biomass yield and grain yield, and one variable related to grain quality, grain protein content. We then analysed the relationship between the relative performance of the mixtures and their functional composition that we characterized with trait means and trait differences on 19 above‐ and below‐ground traits. 3. We found that biomass and grain yield increased by 4% overall in mixtures relative to single varieties, but that mixing effects were non‐significant for grain protein content. The combined effects of trait means and trait differences explained 12%, 17% and 22% of the variability of relative grain yield, biomass yield and grain protein content, respectively, with different traits affecting productivity and grain quality. Clustering varieties into functional groups allowed us to identify the most beneficial associations for multifaceted agronomic performance. 4. Synthesis and applications. Functional traits explained a significant part of the relative agronomic performance of mixtures compared to monocultures (12%–22%, depending on the yield component). They can thus serve as a basis to identify groups of varieties whose combinations are expected to generate positive mixing effects, especially for productivity, and without compromising grain quality. Selection could then target convergence between groups for some traits and divergence between groups for other traits using empirically derived relationships between functional traits and agronomic performance as a guideline.