Dispersal barriers and corridors between habitat patches can strongly affect colonization processes, and therefore patch occupation probability. However, most metapopulation dynamic models assume that heterogeneity in the landscape between patches can be neglected, basing dispersal on the Euclidean (shortest) distance between patches. For heterogeneous landscapes inter-patch distance should take into account the resistance of landscape features to movement, as with the least-cost algorithms implemented in Geographic Information Systems (GIS). In this study, we explore how patch occupancy is sensitive to Euclidean versus a landscape-based distance (least-cost algorithm). We illustrate our method with two metapopulations of the Yellow-bellied Toad (Bombina variegata) in the Rhone plain, Switzerland. The approach allows us to identify which patches are the most sensitive to the inter-patch landscape; i.e., from a conservation point of view, those patches where improving connectivity (e.g., by building vegetated corridors, removing barriers to dispersal) might be a valuable management scenario.