Quantitative methods for the design of reserve networks often select over-dispersed reserves, and consequently a number of species extinctions can be expected in such reserves, especially if unprotected surrounding habitat is lost. A novel approach that deals with this problem is presented by considering factors such as habitat quality and spatial configuration of reserves during the selection process. Species-specific effects of habitat loss and fragmentation, together with habitat composition, are taken into account when computing species probabilities of local occurrence. Sites are then chosen to represent all species with a given target probability. The method is applied to a dataset of butterflies and moths from the Creuddyn Peninsula, North Wales, UK, which includes species with various responses to habitat quality and configuration. The results show that the resulting level of reserve clustering will depend on the number of species for which the spatial configuration plays an important role (at the scale under consideration), and on the pattern and amount of habitat loss that is expected to follow around the reserves. The method represents a step towards taking better into account species persistence when selecting reserve networks in a changing world.