Most population genetic studies in streams infer long-term patterns of gene flow by calculating fixation indices (e.g., F_ST) among sampled populations. In more-recent analytical methods, the need to assign individuals to populations a priori (clustering algorithms) is relaxed, and spatial autocorrelation analysis of allele frequencies (SA) is used to infer finer-scale and potentially short-term dispersal distances. We applied multiple methods to study the population genetic structure of the riverine caddisfly Stenopsyche marmorata (Trichoptera:Stenopsychidae) from 4 adjacent catchments in northeastern Japan. We genotyped larval individuals (N = 532) from 30 sites at 8 polymorphic microsatellite loci. Fixation indices suggested low levels of genetic differentiation among populations (global F_ST = 0.062, p < 0.01), and significant isolation-by-distance (IBD) indicated populations were in drift-migration equilibrium. Bayesian clustering separated S. marmorata into distinct upland (>250 m asl) and lowland populations, with different F_ST values (upland F_ST = 0.048, p < 0.01; lowland F_ST = 0.029, p < 0.01) and significant IBD only among upland populations. Allele frequencies were significantly positively autocorrelated (Moran's I > 0, p < 0.05) at distances up to 18 km along streams and up to 12 km across terrestrial habitat. These values were similar to directly observed flight distance in a single generation for this species in the field. We conclude that the multiple-method approach revealed: 1) unexpected population subdivision between upland and lowland areas that may result from local adaptation, differences in phenology, and historical colonization by multiple lineages; and 2) fine-scale estimates of dispersal that match direct observations of flight and suggest gene flow is more pronounced along water courses in this species.