Landscape genetics is an important framework for investigating the influence of spatial pattern on ecological process. Nevertheless, the standard analytic frameworks in landscape genetics have difficulty evaluating hypotheses about spatial processes in dynamic landscapes. We use a predictive hypothesis-driven approach to quantify the relative contribution of historic and contemporary processes to genetic connectivity. By confronting genetic data with models of historic and contemporary landscapes, we identify dispersal processes operating in naturally heterogeneous and human-altered systems. We demonstrate the approach using a case study of microsatellite polymorphism and indirect estimates of gene flow for a rainforest bird, the logrunner (Orthonyx temminckii). Of particular interest was how much information in the genetic data was attributable to processes occurring in the reconstructed historic landscape and contemporary human-modified landscape. A linear mixed model was used to estimate appropriate sampling variance from nonindependent data and information-theoretic model selection provided strength of evidence for alternative hypotheses. The contemporary landscape explained slightly more information in the genetic differentiation data than the historic landscape, and there was considerable evidence for a temporal shift in dispersal pattern. In contrast, migration rates estimated from genealogical information were primarily influenced by contemporary landscape change. We discovered that landscape heterogeneity facilitated gene flow before European settlement, but contemporary deforestation is rapidly becoming the most important barrier to logrunner dispersal.