Coral reefs are highly complex ecological systems, where multiple processes interact across scales in space and time to create assemblages of exceptionally high biodiversity. Despite the increasing frequency of hierarchically structured sampling programs used in coral‐reef science, little progress has been made in quantifying the relative importance of processes operating across multiple scales. The vast majority of reef studies are conducted, or at least analyzed, at a single spatial scale, ignoring the implicitly hierarchical structure of the overall system in favor of small‐scale experiments or large‐scale observations. Here we demonstrate how α (mean local number of species), β diversity (degree of species dissimilarity among local sites), and γ diversity (overall species richness) vary with spatial scale, and using a hierarchical, information‐theoretic approach, we evaluate the relative importance of site‐, reef‐, and atoll‐level processes driving the fish metacommunity structure among 10 atolls in French Polynesia. Process‐based models, representing well‐established hypotheses about drivers of reef‐fish community structure, were assembled into a candidate set of 12 hierarchical linear models. Variation in fish abundance, biomass, and species richness were unevenly distributed among transect, reef, and atoll levels, establishing the relative contribution of variation at these spatial scales to the structure of the metacommunity. Reef‐fish biomass, species richness, and the abundance of most functional‐groups corresponded primarily with transect‐level habitat diversity and atoll‐lagoon size, whereas detritivore and grazer abundances were largely correlated with potential covariates of larval dispersal. Our findings show that (1) within‐transect and among‐atoll factors primarily drive the relationship between α and γ diversity in this reef‐fish metacommunity; (2) habitat is the primary correlate with reef‐fish metacommunity structure at multiple spatial scales; and (3) inter‐atoll connectedness was poorly correlated with the nonrandom clustering of reef‐fish species. These results demonstrate the importance of modeling hierarchical data and processes in understanding reef‐fish metacommunity structure.