Adaptive population divergence is often driven by divergent natural selection, but can be constrained by the homogenizing effect of gene flow between populations. Indeed, a common pattern in nature is an inverse correlation between the degree of adaptive phenotypic divergence between populations and levels of gene flow between populations. However, there is essentially no experimental data on whether this correlation arises because gene flow constrains adaptation or, conversely, because adaptive divergence causes barriers to gene flow (ecological speciation). Here, I report increased adaptive divergence in cryptic color pattern between a pair of Timema insect populations following an experimental reduction in between-population gene flow. The reduction in gene flow arose due to a natural experiment, and thus was not replicated at a second site. However, temporal replication of the trends among six generations of data, coupled with a lack of increased adaptive divergence for two other population pairs where gene flow was not manipulated (i.e., control sites), argues that the results did not arise by chance. Estimates of dispersal ability and population size further support reduced gene flow, rather than increased genetic drift, as the cause of divergence. Thus, the findings provide experimental evidence that gene flow constrains adaptation in nature.