Résumé: Aim To investigate biotic and abiotic correlates of reef-fish species richness across multiple spatial scales. Location Tropical reefs around the globe, including 485 sites in 109 sub-provinces spread across 14 biogeographic provinces. Time period Present. Major taxa studied 2,523 species of reef fish. Methods We compiled a database encompassing 13,050 visual transects. We used hierarchical linear Bayesian models to investigate whether fish body size, reef area, isolation, temperature, and anthropogenic impacts correlate with reef-fish species richness at each spatial scale (i.e., sites, sub-provinces, provinces). Richness was estimated using coverage-based rarefaction. We also tested whether species packing (i.e., transect-level species richness/m(2)) is correlated with province-level richness. Results Body size had the strongest effect on species richness across all three spatial scales. Reef area and temperature were both positively correlated with richness at all spatial scales. At the site scale only, richness decreased with reef isolation. Species richness was not correlated with proxies of human impacts. Species packing was correlated with species richness at the province level following a sub-linear power function. Province-level differences in species richness were also mirrored by patterns of body size distribution at the site scale. Species-rich provinces exhibited heterogeneous assemblages of small-bodied species with small range sizes, whereas species-poor provinces encompassed homogeneous assemblages composed by larger species with greater dispersal capacity. Main conclusions Our findings suggest that body size distribution, reef area and temperature are major predictors of species richness and accumulation across scales, consistent with recent theories linking home range to species-area relationships as well as metabolic effects on speciation rates. Based on our results, we hypothesize that in less diverse areas, species are larger and likely more dispersive, leading to larger range sizes and less turnover between sites. Our results indicate that changes in province-level (i.e., regional) richness should leave a tractable fingerprint in local assemblages, and that detailed studies on local-scale assemblage composition may be informative of responses occurring at larger scales.