Littoral habitats are commonly defined by environmental properties and the presence of certain dominant species. This conception relies on species’ frequencies and/or abundances, but often omits co-occurrence relationships. In an effort to incorporate this information, we here used presence-absence data from visual censuses of 16 mediolittoral habitats from North-western Mediterranean Sea rocky shores to build co-occurrence networks. These were compared to networks built using species frequency data. Network analyses were done for the entire mediolittoral zone as a unit, and for each habitat individually. For each habitat, network topology revealed central hubs of highly co-occurring species that scored high in degree and eigenvector centrality, essentially corresponding to the most frequent species, both in the mediolittoral zone as a whole and within each of the habitats. Many species, however, co-occurred disproportionately to their frequencies, which highlights the relevance of co-occurrence estimations for habitat descriptions. Some habitats had clear modules as separate habitat subunits. Values of Shannon-Wiener diversity and evenness were well associated with most network metrics, but not with graph density, which helped uncover inhomogeneous species relationships within a same habitat throughout the sampling sites. The habitat features observed mostly supported current phytosociological subdivisions. This stresses the importance of network tools for depicting and describing habitat types, as they provide complementary information to conventional descriptions and may help identify habitat sub-groups, or to define unclassified habitats.