I have studied the consequences of habitat patchiness on the persistence times of deleterious alleles in a random mating population. Results based on computer simulations and supported by analytical approximations suggest that deleterious alleles remain approximately 1/(1−2FST) more generations in the patchy than in a comparable homogeneous population, where 0<FST[les ]0·25 is the fraction of genetic variance due to the sample of families across patches in one generation. In natural populations of Drosophila, therefore, the contribution of deleterious mutants to the genetic variance in fitness might be larger than previously thought. A model of density-dependent viability selection, inspired by the suggestion that deleterious effects can substantially increase when the environment becomes harsher, also gives credence to the analytical results and illustrates that mean persistence times are very sensitive to changes in ecological parameters. If the density dependence model can be taken seriously, there is a clear difficulty in comparing observed and expected levels of genetic variance on the basis of the simplest mutation–selection balance model.