Genetically diverse populations can increase plant resistance to natural enemies. Yet, beneficial genotype pairs remain elusive due to the occurrence of positive or negative effects of mixed planting on plant resistance, respectively called associational resistance or susceptibility. Using genome-wide polymorphism data, we identified genotype pairs with associational resistance to herbivory in the model species Arabidopsis thaliana. With a genome-wide association method named “Neighbor GWAS” and genomic prediction inspired by the Ising model of magnetics we first quantified neighbor interactions among 199 genotypes grown in a randomized block design. These analyses predicted that 823 of the 19,701 candidate pairs could reduce herbivory in mixed planting. We then planted three pairs with the predicted effects in mixtures and monocultures and indeed detected 18–30% reductions in herbivore damage in mixtures. Our study shows the power of genomic prediction to assemble genotype mixtures with positive biodiversity effects.