Species check-lists are helpful to establish Marine Protected Areas (MPAs) and protect local richness, endemicity, rarity, and biodiversity in general. However, such exhaustive taxonomic lists (i.e., true surrogate of biodiversity) require extensive and expensive censuses, and the use of estimator surrogates (e.g., habitats) is an appealing alternative. In truth, surrogate effectiveness appears from the literature highly variable both in marine and terrestrial ecosystems, making it difficult to provide practical recommendations for managers. Here, we evaluate how the biodiversity reference data set and its inherent bias can influence effectiveness. Specifically, we defined habitats by geomorphology, rugosity, and benthic cover and architecture criteria, and mapped them with satellite images for a New-Caledonian site. Fish taxonomic and functional lists were elaborated from Underwater Visual Censuses, stratified according to geomorphology and exposure. We then tested if MPA networks designed to maximize habitat richness, diversity and rarity could also effectively maximize fish richness, diversity, and rarity. Effectiveness appeared highly sensitive to the fish census design itself, in relation to the type of habitat map used and the scale of analysis. Spatial distribution of habitats (estimator surrogate’s distribution), quantity and location of fish census stations (target surrogate’s sampling), and random processes in the MPA design all affected effectiveness to the point that one small change in the data set could lead to opposite conclusions. We suggest that previous conclusions on surrogacy effectiveness, either positive or negative, marine or terrestrial, should be considered with caution, except in instances where very dense data sets were used without pseudo-replication. Although this does not rule out the validity of using surrogates of species lists for conservation planning, the critical joint examination of both target and estimator surrogates is needed for every case study.