Stochastic models for metapopulations that consist of a finite number of local populations are discussed. A stochastic Levins model with two local slates (metapopulation level: a patch is occupied or empty) is compared with a detailed model at the level of local populations. The European badger is taken as an example to illustrate that the stochastic Levins model can form a satisfactory description for populations of real animals, even for those with complex social behaviour. The local population model is a Markov chain model with animals of two sexes and a 1‐year reproduction cycle. Both heuristic and mathematical arguments are given to show that the stochastic Levins model is a sufficiently accurate representation of the complex model. A more general application of this modelling strategy is suggested (or bird and mammal species with a fragmented distribution, yielding a tool for evaluating conservation measures for species threatened by fragmentation. The local population model can then be used to translate changes in population parameters (birth, death, dispersal) into processes on the metapopulation level (local extinction, colonization