A gene diversity analysis was performed using microsatellite loci in order to (i) describe the extent and pattern of population structure in Atlantic salmon (Salmo salar L.) within a river system; (ii) establish the importance of quantifying the signal:noise ratio in accurately estimating population structure; and (iii) assess the potential usefulness of two evolutionary models in explaining within-river population structure from the ecological and habitat characteristics of Atlantic salmon. We found weak, yet highly significant microscale spatial patterning after accounting for variance among temporal replicates within sites. Lower genetic distances were observed among temporal samples at four sampling sites whereas no evidence for temporal stability was observed at the other three locations. The component of genetic variance attributable to either temporal instability and/or random sampling errors was almost three times more important than the pure spatial component. This indicates that not considering signal:noise ratio may lead to an important overestimation of genetic substructuring in situations of weak genetic differentiation. This study also illustrates the usefulness of the member–vagrant hypothesis to generate a priori predictions regarding the number of subpopulations that should compose a species, given its life-history characteristics and habitat structure. On the other hand, a metapopulation model appears better suited to explain the extent of genetic divergence among subpopulations, as well as its temporal persistence, given the reality of habitat patchiness and environment instability. We thus conclude that the combined use of both models may offer a promising avenue for studies aiming to understand the dynamics of genetic structure of species found in unstable environments.