Habitat classes are often used as surrogates to represent or capture species assemblages in the design of spatial conservation strategies, such as multi-use marine protected areas (MPAs). Little research, however, has critically evaluated how well habitat classes can reliably predict species distributions and abundances over scales relevant to spatial planning. In this study, we used hierarchical models to quantify spatial variability in demersal and mid-water fishes at multiple scales to determine whether habitat classes are appropriate surrogates for temperate fishes. Baited remote underwater video systems (BRUVS) and mid-water BRUVS were used to sample fish assemblages in Jervis Bay, NSW, Australia, over rocky reef, seagrass Posidonia australis and unvegetated sediment among locations (>3 km), habitat classes (~400 m) and sites within habitats (~200 m). Each habitat class displayed a distinct assemblage of demersal fish driven by many species and families showing strong habitat associations (e.g. platycephalids and labrids). In contrast, the mid-water fish assemblage and certain demersal families, such as habitat generalists (e.g. sparids), showed no differentiation among habitat class. Considerable variation in the fish assemblage was also observed among locations. Seascape connectivity explained much of this variability, as reefs surrounded by large areas of seagrass harboured a greater abundance and diversity of fishes. Overall, we provide quantitative support for the use of habitat classes as surrogates for most temperate fishes. Spatial planners, however, need to be aware of species-specific relationships with habitat and the importance of seascape patterning when using habitat-based surrogates for MPA design.