Population structure of the threatened long-tailed bat (Chalinolobus tuberculatus) was studied over five summers between 1993 and 1998, in temperate Nothofagus rainforest in Fiordland, New Zealand. Composition of 95 communal groups was sampled and spatial distribution of individually marked bats investigated. Individual C. tuberculatus moved to new roost sites virtually every day. Long term non-random associations among individuals were found by a cluster analysis that revealed three distinct social groups. Groups contained on average 72.0 (± 26.0) (mean ± SD), 99.3 (± 19.0) and 131.7 (± 16.5) marked individuals/year. Collective foraging ranges of the three groups overlapped but roosting occurred in three geographically distinct adjacent areas. Only 1.6% of individuals switched between groups. Non-reproductive females and males switched between groups more often than reproductive females but individuals switched only once or twice during the study and then just for one night. Juveniles of both sexes were associated with their natal group as 1 year-olds and then later when breeding. Social groups were cryptic because foraging ranges of groups overlapped, bats belonging to each group spread over many roosts each day, and these roost sites changed from day to day. Bats moved infrequently between groups, potentially linking the local population assemblages. Future research should explore whether the population is structured in demes. Population structure did not conform to traditional metapopulation models because groups occurred in homogeneous habitat extending over a large geographical area. Conserving bat populations should entail preserving a representative number of subgroups but development of models for predicting minimum number of effective local populations is still required.