Recent theoretical advances have suggested that various forms of balancing selection may promote the evolution of dominance through an increase of the proportion of heterozygote genotypes. We test whether dominance can evolve in the sporophytic self-incompatibility (SSI) system in plants. SSI prevents mating between individuals expressing identical SI phenotypes by recognition of pollen by pistils, which avoids selfing and inbreeding depression. SI phenotypes depend on a complex network of dominance relationships between alleles at the self-incompatibility locus (S-locus). Empirical studies suggest that these relationships are not random, but the exact evolutionary processes shaping these relationships remain unclear. We investigate the expected patterns of dominance under the hypothesis that dominance is a direct target of natural selection. We follow the fate of a mutant allele at the S-locus whose dominance relationships are changed but whose specificity remains unaltered. We show that strict codominance is not evolutionary stable in SSI, and that inbreeding depression due to deleterious mutations linked or unlinked to the S-locus exerts strong constraints on changes in relative levels of dominance in pollen and pistil. Our results provide a general adaptive explanation for most patterns of dominance relationships empirically observed in natural plant populations.