To assess how the decimation of large vertebrates by hunting alters recruitment processes in a tropical forest, we compared the sapling cohorts of two structurally and compositionally similar forests in the Rio Manu floodplain in southeastern Peru. Large vertebrates were severely depleted at one site, Boca Manu (BM), whereas the other, Cocha Cashu Biological Station (CC), supported an intact fauna. At both sites we sampled small (> or =1 m tall, <1 cm dbh) and large (> or =1 cm and <10 cm dbh) saplings in the central portion of 4-ha plots within which all trees > or =10 cm dbh were mapped and identified. This design ensured that all conspecific adults within at least 50 m (BM) or 55 m (CC) of any sapling would have known locations. We used the Janzen-Connell model to make five predictions about the sapling cohorts at BM with respect to CC: (1) reduced overall sapling recruitment, (2) increased recruitment of species dispersed by abiotic means, (3) altered relative abundances of species, (4) prominence of large-seeded species among those showing depressed recruitment, and (5) little or no tendency for saplings to cluster closer to adults at BM. Our results affirmed each of these predictions. Interpreted at face value, the evidence suggests that few species are demographically stable at BM and that up to 28% are increasing and 72% decreasing. Loss of dispersal function allows species dispersed abiotically and by small birds and mammals to substitute for those dispersed by large birds and mammals. Although we regard these conclusions as preliminary, over the long run, the observed type of directional change in tree composition is likely to result in biodiversity loss and negative feedbacks on both the animal and plant communities. Our results suggest that the best, and perhaps only, way to prevent compositional change and probable loss of diversity in tropical tree communities is to prohibit hunting.