In most metapopulation models dispersal is assumed to be a fixed species-specific trait, but in reality dispersal abilities are highly sensitive to various selective pressures. Strict isolation of a metapopulation, which precludes any influx of immigrants (and their genes) from outside and makes it impossible for emigrants to reach other localities with suitable habitat, thus reducing fitness benefits of long-distance dispersal to zero, may be expected to impose strong selection against dispersal. We tested the above prediction by comparing dispersal parameters derived with the Virtual Migration model for isolated and non-isolated metapopulations of two species of large blue Maculinea (= Phengaris) butterflies, surveyed with intensive mark-recapture. Mortality during dispersal was found to be twice (in M. teleius) to five times higher (in M. arion) in isolated metapopulations. Isolation also resulted in significantly reduced dispersal distances in isolated metapopulations, with the effect being particularly strong in M. arion females. Apart from its evolutionary and ecological consequences, dispersal depression in isolated butterfly metapopulations implied by our results has serious conservation implications. It provides a clear argument against using parameter values obtained in a different environmental setting in modelling applications, e.g., Population Viability Analyses or environmental impact assessment. Furthermore, it underlines the importance of establishing well-connected networks of suitable habitats prior to species release in areas where reintroductions are planned.