Most studies of rare and endangered species are based on work carried out within one generation, or over one to a few generations of the study organism. We report the results of a study that spans 30 generations (years) of the entire natural range of a butterfly race that is endemic to 35 km(2) of north Wales, UK. Short-term studies (surveys in single years and dynamics over 4 years) of this system led to the prediction that the regional distribution would be quite stable, and that colonization and extinction dynamics would be relatively unimportant. However, a longer-term study revealed unexpectedly high levels of population turnover (local extinction and colonization), affecting 18 out of the 20 patches that were occupied at any time during the period. Modelling the system (using the 'incidence function model' (IFM) for metapopulations) also showed higher levels of colonization and extinction with increasing duration of the study. The longer-term dynamics observed in this system can be compared, at a metapopulation level, with the increased levels of variation observed with increasing time that have been observed in single populations. Long-term changes may arise from local changes in the environment that make individual patches more or less suitable for the butterfly, or from unusual colonization or extinction events that take metapopulations into alternative states. One implication is that metapopulation and population viability analyses based on studies that cover only a few animal or plant generations may underestimate extinction threats.