Spatial and temporal constraints on dispersal explain the absence of species from areas with potentially suitable conditions. Previous studies have shown that post-glacial recolonization has shaped the current ranges of many species, yet it is not completely clear to what extent interspecific differences in range size depend on different dispersal rates. The inferred boundaries of glacial refugia are difficult to validate, and may bias spatial distribution models (SDMs) that consider post-glacial dispersal constraints. We predicted the current distribution of 12 Caucasian forest plants and animals, factoring in the effective geographical distance from inferred glacial refugia as an additional predictor. To infer glacial refugia, we tested the transferability of the current SDMs based on the distribution of climatic variables, and projected the most transferable ones onto two climate scenarios simulated for the Last Glacial Maximum (LGM). We then calculated least-cost distances from the inferred refugia, using elevation as a friction surface, and recalculated the current SDMs incorporating the distances as an additional variable. We compared the predictive powers of the initial with the final SDMs. The palaeoclimatic simulation that best matched the distribution of species was assumed to represent the closest fit to the true palaeoclimate. SDMs incorporating refugial distance performed significantly better for all but one studied species, and the Model for Interdisciplinary Research on Climate (MIROC) climatic simulation provided a more convincing pattern of the LGM climate than the Community Climate System Model (CCSM) simulation. Our results suggest that the projection of suitable habitat models onto past climatic conditions may yield realistic boundaries of glacial refugia, and that the current distribution of forest species in the study region is strongly associated with locations of former refugia. We inferred six major forest refugia throughout western Asia: (1) Colchis; (2) western Anatolia; (3) western Taurus; (4) the upper reaches of the Tigris River; (5) the Levant; and (6) the southern Caspian basin. The boundaries of the modelled refugia were substantially broader than the refugia boundaries inferred solely from pollen records. Thus, our method could be used to: (1) improve models of current species distributions by considering the dispersal histories of the species; and (2) validate alternative reconstructions of palaeoclimate with current distribution data. © 2011 The Linnean Society of London, Biological Journal of the Linnean Society, 2012, 105, 231–248.