We monitored a close-spaced grid of 289 seed traps in 1.44 ha for 8.4 yr in an Amazonian floodplain forest. In a tree community containing hundreds of species, a median of just three to four species of tree seeds falls annually into each 0.5-m2 establishment site. The number of seed species reaching a given site increased linearly with time for the duration of the monitoring period, indicating a roughly random arrival of seed species in a given site-year. The number of seed species captured each year over the entire grid ranged from one-third to one-half of the total captured over the 8.4 yr of monitoring, revealing a substantial temporal component of variation in the seed rain. Seed rain at the 0.5-m2 scale displayed extreme spatial variability when all potentially viable seeds were tallied, whereas the rain of dispersed seeds was scant, more nearly uniform, and better mixed. Dispersal limitation, defined as failure of seeds to reach establishment sites, is ≥99% per year for a majority of species, explaining why seed augmentation experiments are often successful. Dispersal limitation has been evoked as an explanation for distance-dependent species turnover in tropical tree communities, but that interpretation contrasts with the fact that many Amazonian tree species possess large geographical ranges that extend for hundreds or thousands of kilometers. A better understanding of the processes that bridge the gap between the scales of seedling establishment and the regulation of forest composition will require new methodologies for studying dispersal on scales larger than those yet achieved.