Populations of predominantly selfing plant species often show spatial genetic structure but little is known whether epistatic gene interactions are spatially structured. To detect a possible epistatic effect and a spatial scale at which it operates, we created artificial crosses between plants spanning a range of fixed distances from 1 to 400 m in three populations of wild barley. The self-pollinated and crossed progeny (F 1 ) and two generations of segregated progeny (F 2 and F 3 ) were tested in experimentally simulated population environments for relative performance (RP). The measured fitness traits included number of seeds, total seed weight and seed germination. For any of these traits, there was no association between RP of F 1 , F 2 and F 3 plants and either pairwise kinship coefficients or crossing distance. In contrast, in all three populations, we found lower seed viability of outcrossed as compared with self-pollinated genotypes in the first generation of segregation. However, in the F 3 generation this outbreeding effect disappeared in the two populations and greatly decreased in the third population. For seed production, heterosis in F 1 and outbreeding depression in F 2 were observed only in the population with unusually high number of heterozygotes. Our findings support the view that in selfing species a spatial mosaic of various locally abundant genotypes represents not randomly fixed combinations of alleles but the co-adapted gene complexes that were sieved by selection, while heterozygotes are characteristic for the transient phase of this process, when segregation and purging of maladaptive genotypes have not yet occurred.