In long-lived animals, adult survival is among the most important determinants of population dynamics. Although it may show considerable variation both in time and among populations and sites, a single survival estimate per species is often used in comparative evolutionary studies or in conservation management to identify threatened populations. We estimated adult survival of the isabelline serotine bat Eptesicus isabellinus using capture-recapture data collected on six maternity colonies scattered over a large area (distance 8-103 km) during periods varying from 8 to 26 years. We modelled temporal and inter-colony variations as random effects in a Bayesian framework and estimated mean annual adult survival of females on two scales and a single survival value across all colonies. On a coarse scale, we grouped colonies according to two different habitat types and investigated the effect on survival. A difference in adult survival was detected between the two habitat types [posterior mean of annual survival probability 0.71; 95% credible interval (CI) 0.51-0.86 vs. 0.60; 0.28-0.89], but it was not statistically supported. On a fine scale, survival of the six colonies ranged between 0.58 (95% CI 0.23-0.92) and 0.81 (0.73-0.88), with variation between only two colonies being statistically supported. Overall survival was 0.72 (95% CI 0.57-0.93) with important inter-colony variability (on a logit scale 0.98; 95% CI 0.00-8.16). Survival varied temporally in a random fashion across colonies. Our results show that inference based solely on single colonies should be treated with caution and that a representative unbiased estimate of survival for any species should ideally be based on multiple populations.