The social structure of an Old World tent-making bat Cynopterus sphinx (Megachiroptera), was investigated in western India. A combination of census and mark–recapture data over 2 years (1996–98) was used to infer the form of the mating system, compositional stability of social groups and mode of new social group formation. The breeding population of C. sphinx was subdivided into diurnal roosting colonies, each of which contained one to five discrete roosting groups and often one or more solitary bats in adjacent roosts. Bats most frequently roosted in stem tents constructed in the flower/fruit clusters of the kitul palm Caryota urens. Temporal variation in social structure was assessed using visual census data for a subset of the study population over 3 years (1995–98) spanning six consecutive reproductive periods. The sex and age composition of diurnal roosting groups indicated a polygynous harem-forming mode of social organization, as groups invariably contained a single adult male, 1–37 reproductive females and their dependent young (n = 33 harems). Harem size averaged 6.1 adults in the wet season (n = 19, SD = 3.5) and 13.6 adults in the dry season (n = 14, SD> = 8.5). The same harem social configuration was maintained year-round, despite a high degree of synchrony and seasonality in the timing of reproduction. Juveniles of both sexes dispersed after weaning and sexually immature bats were never present in harems at the time of parturition. Adult females often remained associated as roostmates from one parturition period to the next, and group cohesion was unaffected by turnover of harem males. Adult females frequently transferred among roosts within the same colony, and harems underwent periodic fissions and fusions. The founding of new harems most often resulted from the fissioning of previously cohesive harems within the same colony. However, some harems contained disproportionate numbers of yearling females, indicating that new groups are also founded by nulliparous females of the same age cohort. A significant degree of heterogeneity in age composition among harems was revealed in the 1998 dry season, but was unrelated to age-stratification of tent roosts. Although formation of new harems may be non-random with respect to age composition of the founders, founding events are not restricted to newly created tents and often involve recolonization of previously occupied roosts.