There is ongoing debate regarding the effects of seed dispersal and seedling establishment on species composition and dynamics in diverse forests. The stakes in this debate are high, because neutral theories of coexistence depend upon a strong role of dispersal, whereas niche-differentiation theories suggest that establishment may be stronger. We added seeds of eight species at five levels of density to examine the effects of seed dispersal on seedling stem density in a diverse tropical forest. Separately, we sowed five levels of species richness (controlling for density, and randomizing composition) to test the effects of dispersal on seedling diversity and species composition. We censused the focal seedlings every six months for two years. At each census, we assessed the biotic neighborhood, canopy openness, and soil moisture to estimate the effects of establishment conditions on seedling recruitment. Stepwise multiple regression indicated that dispersal treatment explained more variance in seedling density and diversity than did establishment conditions, though the effects of dispersal weakened over the two years following sowing. Mantel tests indicated that dispersal treatments, but not post-dispersal establishment conditions, were positively correlated with focal seedling species composition. Furthermore, overall (focal + non-focal) seedling composition was more tightly correlated with dispersal treatment than with establishment conditions. This is the first study to show that seed dispersal can determine not only seedling density, but also diversity and species composition. Our results support a view that dispersal may more strongly influence the species composition of young tropical trees than establishment conditions, consistent with predictions from neutral models.