The "intermediate disturbance hypothesis," which postulates maximum diversity at intermediate regimes of disturbance, has never been clearly proved to apply to species-rich tropical forest tree communities and to local-scale canopy disturbances that modify light environments. This hypothesis was tested on a sample of 17,000 trees in a Guianan forest, 10 years after a silvicultural experiment that added to natural treefall gaps a wide range of disturbance intensities. Species richness, standardized to eliminate density effects, peaked at intermediate disturbance levels, particularly when disturbance intensity was estimated through the percentage of stems of strongly light-dependent species.