Models of costs and benefits of dormancy (D) predict that the evolutionarily stable strategy in long-term stable environments is for non-dormancy (ND), but this prediction remains to be tested empirically. We reviewed seed traits of species in the climatically buffered, geologically stable and nutrient-impoverished campo rupestre grasslands in Brazil to test the hypothesis that ND is favoured over D. We examined the relative importance of life-history traits and phylogeny in driving the evolution of D and assessed seed viability at the community level.