Crop losses to pests, or crop gains to diversification: what matters most? This is a pivotal question for the acceptance of crop diversification solutions by farmers and breeders. At the intraspecific level in particular, the magnitude of beneficial effects of mixing varieties on crop yield compared to disease attacks and weed invasion in the absence of pesticide remains questioned. Here, we quantified for the first time the relative importance of different types of biotic interactions – crop-crop interactions, crop-fungi interactions and crop-weed interactions, on grain yields in a region of the world where alternative agricultural practices are urgently needed to face the multiple pressures induced by global changes. We evaluated the performance of ten upland rice varieties of Madagascar highlands using a field trial where all varieties were grown in pure stands and in bi-variety mixtures. Grain production and disease attacks by Pyricularia oryzea were assessed in all varieties and plots. Weed biomass was recorded in all plots. We also measured plant traits involved in plant-plant interactions - plant height and flowering time in particular - for all plots and varieties to identify generic assembly rules for optimizing the performance of varietal mixtures. We found strong positive effects of rice mixture on both crop yield and pathogen resistance while weed biomass remained constant and high whatever the experimental condition. When statistically controlling for one pressure, we demonstrated that the beneficial effects of crop-crop interactions on yield largely overcame the negative effects of crop-fungi and crop-weed interactions. The average height of varieties in mixtures as well as their height plasticity had strong predictive power of crop yields, confirming the great potential of trait-based approaches to design innovative cropping system. Overall, the predominant, positive effect of varietal mixtures over disease and weed pressures represents a key argument for crop diversification.