1. The consequences of biodiversity loss for ecosystem services largely depend on the functional identities of extirpated species. However, poor descriptions of spatial patterns of community functional composition across landscapes hamper accurate predictions, particularly in highly diverse tropical regions. Therefore, understanding how community functional composition varies across environmental gradients remains an important challenge. 2. We sampled 15 functional traits in 800 Neotropical tree species across 13 forest plots representative of the broad climatic and soil gradients encompassed by three widespread lowland forest habitats (terra firme forests on clay-rich soils, seasonally flooded forests and white-sand forests) at opposite ends of Amazonia (Peru and French Guiana). We combined univariate and multivariate approaches to test the magnitude and predictability of environmental filtering on community leaf and wood functional composition. 3. Directional shifts in community functional composition correlated with environmental changes across the 13 plots, with denser leaves, stems and roots in forests occurring in environments with limited water and soil-nutrient availability. Critically, these relationships allowed us to accurately predict the functional composition of 61 additional forest plots from environmental data alone. 4. Synthesis. Environmental filtering consistently shapes the functional composition of highly diverse tropical forests at large scales across the terra firme, seasonally flooded and white-sand forests of lowland Amazonia. Environmental factors drive and allow the prediction of variation in community functional composition among habitat types in Amazonian forests.