The pattern of geographic variation in tree-climbing ability of Peromyscus maniculatus was used as a natural experiment to examine the interaction of natural selection, genetic drift, and gene flow. The divergence in climbing behavior among lab-reared mice derived from adults trapped in forest, woodland, and desert habitats was compared with a series of adaptive and non-adaptive hypotheses of evolutionary change. Natural selection was predicted to produce better climbers in forests and woodlands than in deserts whereas divergence due to genetic drift was expected to be independent of habitat type. Gene flow between neighboring habitats was predicted to reduce differentiation in climbing traits. Tree-climbing ability was measured by determining the maximum diameter artificial trunk (rod) that a mouse could climb in escaping from a lighted, confined area. Larger diameter rod scores reflect better climbing abilities (Horner 1954). Comparisons of mean rod climbing scores between subspecies and among forest, woodland, and desert habitats reveal that P. m. rufinus, sampled from forest and woodland, is a better climber than P. m. sonoriensis, sampled from woodland and desert habitats. This is consistent with the hypothesis that natural selection has produced subspecies level adaptation in climbing behavior. However, the climbing ability of P. m. sonoriensis sampled from woodland habitats on isolated mountaintops, although slightly divergent from populations in adjacent desert scrub habitats, has not evolved in response to natural selection to the degree expected from the observed subspecies level adaptation. Additionally, populations of unknown subspecific status sampled from desert grassland habitat, adjacent to woodland P. m. rufinus, have climbing abilities that are not significantly different from woodland forms. Thus, evolution in certain populations is constrained. If gene flow from desert populations into woodland mountaintop populations is constraining evolution, then mountaintop populations should have high trait variances. Analysis of the within population variance does not support this hypothesis. Other lines of evidence that indicate gene flow is low or moderate are reviewed. In conclusion, adaptation to local habitats is constrained, perhaps by restrictive genetic correlations and/or lack of sufficient time to respond to natural selection for climbing ability. As a result, long periods of consistent selection are necessary to produce the pattern of subspecific adaptation in climbing behavior. (Abstract shortened with permission of author.)