1 We hypothesized that severe drought affects the structure of tropical forests by favouring seedlings of some species or groups at the expense of others. To test this hypothesis, we irrigated naturally occurring woody seedlings during an El Niño-related drought in seasonal moist tropical forest on Barro Colorado Island, Panama. We predicted that irrigated plots would retain greater species richness than control plots, and that the benefits of irrigation would increase with the abundance of: trees relative to lianas, wet-forest seedlings relative to dry-forest seedlings, and rare species relative to common species. We also hypothesized that the strength of this filter would increase with increased moisture limitation, predicting that the benefits of irrigation would increase with seedling density and light availability, and decrease with mean seedling age. 2 Irrigation did reduce species losses, but not by limiting the loss of drought-sensitive species as predicted. Instead, mortality in irrigated plots was density dependent, whereas species losses in control plots were well predicted by random thinning, suggesting that density dependence weakened as abiotic stress increased. 3 Irrigation increased seedling growth, but did not affect seedling mortality. Contrary to our predictions, irrigation increased growth in plots dominated by dry-forest species relative to those dominated by wet-forest species, suggesting that dry-forest seedlings either occur in moisture-limited microsites or are more able to utilize dry-season precipitation. The strength of the filter did increase with potential moisture limitation, as irrigation increased seedling growth more in higher light environments. 4 Annual precipitation has declined over much of the humid tropics during the 20th century. Our results suggest that this trend may reduce tropical forest diversity by weakening density-dependent mechanisms that maintain diversity. In addition, plots dominated by dry-forest species experienced higher growth in response to irrigation and also far lower dry-season mortality relative to plots dominated by wet-forest species. While we cannot disentangle the effects of microsite from species composition, these results suggest that dry-forest species may benefit from any increase in dry season length or severity. 5 Research conducted during ‘normal’ conditions may overlook the impact of severe events and thus fail to identify critical mechanisms structuring ecological communities.