Efficient biodiversity monitoring programs are essential to assess the ecological status of marine ecosystems. In the context of the European Marine Strategy Framework Directive (MFSD), coastal rocky reef fish assemblages are most often surveyed using underwater visual census (UVC). While UVC is well-suited to monitor conspicuous fish, it underestimates hidden, vagrant and/or elusive fishes. Hence, complementary methods are needed to survey the whole fish assemblage. Environmental DNA (eDNA) metabarcoding is increasingly used in this context. Here, we evaluated the effectiveness of eDNA metabarcoding collected from water to provide sound and complementary diversity data to UVC in support of the MFSD. Both methods were used to examine coastal rocky reef fish assemblages in one bay of Brittany (France), along an inshore-offshore gradient and during two seasons. eDNA metabarcoding was carried out following different water sampling strategies: small water volumes (2L) from stationary sampling at the surface and bottom and large water volume (30L) filtered at the surface while moving in close vicinity of the sampling site. A total of 93 (including 57 species) and 33 (including 27 species) taxa were detected with eDNA metabarcoding and UVC, respectively. Eleven species were only recorded by UVC, a result explained for nine of them by a lack of taxonomic resolution of the markers used (12S and 16S) and/or the unavailability of reference sequences. eDNA metabarcoding allows to recover species expected to be present, such as demersal rocky specialist fishes, at a similar species richness level than UVC (14 species for UVC and 12 species for eDNA). It however also unveiled other taxa (pelagic, bentho-pelagic and others demersal fishes) inhabiting the bay and rarely reported through UVC (18 species for UVC and 45 species for eDNA). Despite these differences in species identified by UVC vs. eDNA, both survey methods were consistent in characterizing spatiotemporal variabilities of fish assemblages, highlighting strong site fidelity of eDNA to the source communities. In a well-mixed water column, the eDNA signal was very homogeneous, allowing the use of the cost-effective small water volume eDNA sampling method. eDNA metabarcoding is thus particularly promising to study coastal rocky reef assemblages, and when coupled with UVC, it improves monitoring surveys.