The effects of habitat fragmentation on diverse tropical tree communities are poorly understood. Over a 20-year period we monitored the density of 52 tree species in nine predominantly successional genera (Annona, Bellucia, Cecropia, Croton, Goupia, Jacaranda, Miconia, Pourouma, Vismia) in fragmented and continuous Amazonian forests. We also evaluated the relative importance of soil, topographic, forest dynamic, and landscape variables in explaining the abundance and species composition of successional trees. Data were collected within 66 permanent 1-ha plots within a large (approximately 1000 km2) experimental landscape, with forest fragments ranging from 1 to 100 ha in area. Prior to forest fragmentation, successional trees were uncommon, typically comprising 2-3% of all trees (> or =10 cm diameter at breast height [1.3 m above the ground surface]) in each plot. Following fragmentation, the density and basal area of successional trees increased rapidly. By 13-17 years after fragmentation, successional trees had tripled in abundance in fragment and edge plots and constituted more than a quarter of all trees in some plots. Fragment age had strong, positive effects on the density and basal area of successional trees, with no indication of a plateau in these variables, suggesting that successional species could become even more abundant in fragments over time. Nonetheless, the 52 species differed greatly in their responses to fragmentation and forest edges. Some disturbance-favoring pioneers (e.g., Cecropia sciadophylla, Vismia guianensis, V. amazonica, V. bemerguii, Miconia cf. crassinervia) increased by >1000% in density on edge plots, whereas over a third (19 of 52) of all species remained constant or declined in numbers. Species responses to fragmentation were effectively predicted by their median growth rate in nearby intact forest, suggesting that faster-growing species have a strong advantage in forest fragments. An ordination analysis revealed three main gradients in successional-species composition across our study area. Species gradients were most strongly influenced by the standlevel rate of tree mortality on each plot and by the number of nearby forest edges. Species-composition also varied significantly among different cattle ranches, which differed in their surrounding matrices and disturbance histories. These same variables were also the best predictors of total successional-tree abundance and species richness. Successional-tree assemblages in fragment interior plots (>150 m from edge), which are subjected to fragment area effects but not edge effects, did not differ significantly from those in intact forest, indicating that area effects per se had little influence on successional trees. Soils and topography also had little discernable effect on these species. Collectively, our results indicate that successional-tree species proliferate rapidly in fragmented Amazonian forests, largely as a result of chronically elevated tree mortality near forest edges and possibly an increased seed rain from successional plants growing in nearby degraded habitats. The proliferation of fast-growing successional trees and correlated decline of old-growth trees will have important effects on species composition, forest dynamics, carbon storage, and nutrient cycling in fragmented forests.