A hierarchical sampling design, spanning five orders of magnitude (from 10s of metres to 100s of kilometres) was created in order to quantify the multiscale spatial variability of visually censused rocky reef ish assemblages in the western Mediterranean Sea. Specifically, we tested the hypotheses that species abundance and the biomass of reef fish populations is higher within than outside marine reserves (Cabo de Gata Nijar, Aguilas, Cabo de Palos, Isla Grosa, Cabrera, Western-Mallorca and Columbretes), and that a north-to-south geographical gradient of these variables exists. We also explored the relationship between the fish assemblage and habitat structure, as an environmental factor likely to account for an important part of the observed variability. The mixed analyses of variance revealed that total abundance and biomass, species richness and abundance and biomass of several target species reached higher average values within marine reserves. Nevertheless, some non-protected localities (e.g. Aguilas) harboured richer and more abundant fish assemblages than some marine reserves. In addition, regional variation, attributable to differences in local carrying capacity and hydroclimatic conditions, are also shown across the studied area. Moreover, the studied assemblage is patchy at small and/or intermediate spatialscales, considering both assemblage descriptors (total abundance and biomass, species richness), and the abundance and biomass of fish species and spatial categories.Detected patterns were different depending on the species and assemblage variables analysed. Differences in habitat structure account for a significant proportion of total variability of the studied variables, and are likely to be responsible for a large part of the observed differences, especially at small-to-intermediate spatial scales. Other factors-spatial variability in larval distribution, settlement and/or post-settlement survival- are discussed in order to explain the observed differences.We concluded that causes of the observed patchiness of Mediterranean reef fish assemblages are probably multiple. Long-term, multi-scale spatial and temporal monitoring actions, as well as process-oriented manipulative experiments are urgently needed in order to ascertain the relative importance of each factor.