Movements between habitat patches in a patchy population of the butterfly Boloria aquilonaris were monitored using capture-mark-recapture methods during three successive generations. For each data set, the inverse cumulative proportion of individuals moving 100 m distance classes was fitted to the negative exponential function and the inverse power function. In each case, the negative exponential function provided a better fit than the inverse power function. Two dispersal kernels were generated using both negative exponential and inverse power functions. These dispersal kernels were used to predict movements between 14 suitable sites in a landscape of 220 km2. The negative exponential function generated a dispersal kernel predicting extremely low probabilities for movements exceeding 1 km. The inverse power function generated probabilities predicting that between site movements were possible, according to metapopulation size. CMR studies in the landscape revealed that long distance movements occurred at each generation, corresponding to predictions of the inverse power function dispersal kernel. A total of 26 movements between sites (up to 13 km) were detected, together with recolonisation of empty sites. The spatial scale of the metapopulation dynamics is larger than ever reported on butterflies and long distance movements clearly matter to the persistence of this species in a highly fragmented landscape.