Seed dormancy is expected to provide ecological advantages by adjusting germination to the favorable growth period. However, many species produce nondormant seeds, particularly in wet tropical forests, a biogeographic pattern that is not well accounted for in current models. We hypothesized that the global distribution of dormant seeds derives from their adaptive value in predictably fluctuating (i.e. seasonal) environments. However, the advantage conferred by dormancy might ultimately depend on other seed attributes, particularly size. This general model was tested within a phylogenetically informed framework using a data set comprising > 216 000 world-wide observations of Fabaceae, spanning three orders of magnitude in seed size and including both dormant and nondormant seeds. Our results confirmed our hypothesis: nondormant seeds can only evolve in climates with long growing seasons and/or in lineages that produce larger seeds. Conversely, dormancy should be evolutionarily stable in temperate lineages with small seeds. When the favorable season is fleeting, seed dormancy is the only adaptive strategy. Based on these results, we predict that, within a given lineage, taxa producing larger, nondormant seeds will necessarily predominate in aseasonal environments, while plants bearing small, dormant seeds will be dominant under short growing seasons.