Phylogeography can reveal evolutionary processes driving natural genetic-geographical patterns in biota, providing an empirical framework for optimizing conservation strategies. The long-term population history of a rotting-log-adapted giant springtail (Collembola) from montane southeast Australia was inferred via joint analysis of mitochondrial and multiple nuclear gene genealogies. Contemporary populations were identified using multilocus nuclear genotype clustering. Very fine-scale sampling combined with nested clade and coalescent-based analyses of sequences from mitochondrial cytochrome oxidase I and three unlinked nuclear loci uncovered marked population structure, deep molecular divergences, and abrupt phylogeographical breaks over distances on the order of tens of kilometres or less. Despite adaptations that confer low mobility, rare long-distance gene flow was implicated: novel computer simulations that jointly modelled stochasticity inherent in coalescent processes and that of DNA sequence evolution showed that incomplete lineage sorting alone was unable to explain the observed spatial-genetic patterns. Impacts of Pleistocene or earlier climatic cycles were detected on multiple timescales, and at least three putative moist forest refuges were identified. Water catchment divisions predict phylogeographical patterning and present-day population structure with high precision, and may serve as an excellent surrogate for biodiversity indication in sedentary arthropods from topographically heterogeneous montane temperate forests.