The genetic structure of brown trout (Salmo trutta) populations inhabiting rivers on the island of Bornholm in the Baltic Sea was studied on a spatial and temporal scale. Low water levels in the rivers during the summer period are assumed to have a significant impact on the persistence of local populations, possibly resulting in a metapopulation structure. Extinctions may, however, also be buffered by a remnant strategy, whereby juveniles escape to river outlets during periods of drought. We compared polymorphism at seven microsatellite DNA loci in contemporary and past samples collected from 1944 to 1997. A principal component analysis, a hierarchical gene diversity analysis and assignment tests showed that the genetic composition of populations was not temporally stable, and that temporal genetic differentiation was much stronger than spatial differentiation. Genetic variability was high and stable over time. Effective population sizes (Ne) and migration rate (m) were estimated using a maximum-likelihood-based implementation of the temporal method. Ne estimates were low (ranging from 8.3 to 22.9) and estimates of m were high (between 0.23 and 0.99), in contrast to other Danish trout populations inhabiting larger and more environmentally stable rivers (Ne between 39.2 and 289.9 and m between 0.01 and 0.09). Thus, the observed spatio-temporal patterns of genetic differentiation can be explained by drift in small persisting populations, where levels of genetic variation are maintained by strong gene flow. However, observations of rivers devoid of trout suggested that population turnover also takes place. We suggest that Bornholm trout represent a metapopulation where the genetic structure primarily reflects strong drift and gene flow, combined with occasional extinction–recolonization events.