Abstract The deep reef refugia hypothesis (DRRH) predicts that deep reef ecosystems may act as refugium for the biota of disturbed shallow waters. Because deep reefs are among the most understudied habitats on Earth, formal tests of the DRRH remain scarce. If the DRRH is valid at the community level, the diversity of species, functions, and lineages of fish communities of shallow reefs should be encapsulated in deep reefs. We tested the DRRH by assessing the taxonomic, functional, and phylogenetic diversity of 22 Brazilian fish communities between 2 and 62 m depth. We partitioned the gamma diversity of shallow (<30 m) and deep reefs (>30 m) into independent alpha and beta components, accounted for species’ abundance, and assessed whether beta patterns were mostly driven by spatial turnover or nestedness. We recorded 3,821 fishes belonging to 85 species and 36 families. Contrary to DRRH expectations, only 48% of the species occurred in both shallow and deep reefs. Alpha diversity of rare species was higher in deep reefs as expected, but alpha diversity of typical and dominant species did not vary with depth. Alpha functional diversity was higher in deep reefs only for rare and typical species, but not for dominant species. Alpha phylogenetic diversity was consistently higher in deep reefs, supporting DRRH expectations. Profiles of taxonomic, functional, and phylogenetic beta diversity indicated that deep reefs were not more heterogeneous than shallow reefs, contradicting expectations of biotic homogenization near sea surface. Furthermore, pairwise beta‐diversity analyses revealed that the patterns were mostly driven by spatial turnover rather than nestedness at any depth. Conclusions. Although some results support the DRRH, most indicate that the shallow‐water reef fish diversity is not fully encapsulated in deep reefs. Every reef contributes significantly to the regional diversity and must be managed and protected accordingly.