From 1989 to 1999, fish assemblages were sampled by a non desctructive visual census method in Posidonia oceanica seagrass beds of the Scandola marine reserve. Censuses were carried out at 13 stations belonging to 3 sectors: outside the reserve (no protection), non integral reserve (professional fishermen allowed with special permit) and integral reserve (no take area). The species richness by transect and mean density exhibited yearly fluctuations, but over the decade, were less variable in protected sectors that in the non protected sector. The level of variation of total fish assemblage density is lower than those of the main families (Labridae, Sparidae and Serranidae). The long-lived species (e.g., Serranus scriba) displayed lower density variations, i.e., higher inertia, that short-lived species (e.g., Symphodus ocellatus). The resilience of the assemblage (buffer effect) was not affected by spatial heterogeneity, but is related to the level of protection. The higher densities of predator in protected seagrass beds (e.g., Serranidae) and in bordering rocky areas could involve a higher predation pressure upon Posidonia oceanica fish assemblage. The buffer effect may be linked to this increase in predation pressure and fishing prohibition. While long-term data are scarce, available information suggests that, due to protection inside marine protected area, fish assemblages of seagrass beds might be relatively resilient to environmental perturbations. This study highlights the necessity of long-term surveys to support pertinent conclusions on the dynamics of coastal fish assemblages.