Aim: How historical and contemporary eco-evolutionary processes shape the patterns of genetic diversity and differentiation across species’ distribution range remains an open question with strong conservation implications. Focusing on the orange stony coral, Astroides calycularis, we (a) characterized the pattern of neutral genetic diversity across the distribution range; (b) gave insights into the underlying processes; and (c) discussed conservation implications with emphasis on a national park located on a hotspot of genetic diversity. Location: South Mediterranean Sea and Zembra National Park. Methods: We combined new data from 12 microsatellites in 13 populations located in the Centre and in the Western Periphery of the distribution range with a published dataset including 16 populations from the Western and Eastern Peripheries. We analysed the relationship among parameters of genetic diversity (He, Ar(g)) and structure (population-specific FST) and two measures of geographic peripherality. We compared two estimators of pairwise genetic structure (GST, DEST) across the distribution range. The evolutionary and demographic history of the populations following the Last Glacial Maximum was reconstructed using approximate Bayesian computations and maximum-likelihood analyses. We inferred the contemporary connectivity among populations from Zembra National Park and with the neighbouring area of Cap Bon. Results: We demonstrate a decrease in genetic diversity and an increase in genetic differentiation from the Centre to the Eastern and Western Peripheries of the distribution range. Populations from Zembra show the highest genetic diversity reported in the species. We identified a spillover effect towards Cap Bon. Main conclusions: The patterns of genetic diversity and differentiation are most likely explained by “the postglacial range expansion hypothesis” rather than the “central–peripheral hypothesis.” Enforcement of conservation measures should be considered to protect this genetic diversity pattern, in particular when considering the low effective population size inferred at many sites.