1. Metacommunity research relies largely on proxies for inferring the effect of dispersal on local community structure. Overland and watercourse distances have been typically used as such proxies. A good proxy for dispersal should, however, take into account more complex landscape features that can affect an organism's movement and dispersal. The cost distance approach does just that, allowing determining the path of least resistance across a landscape. 2. Here, we examined the distance decay of assemblage similarity within a subarctic stream insect metacommunity. We tested whether overland, watercourse and cumulative cost distances performed differently as correlates of dissimilarity in assemblage composition between sites. We also investigated the effect of body size and dispersal mode on metacommunity organization. 3. We found that dissimilarities in assemblage composition correlated more strongly with environmental than physical distances between sites. Overland and watercourse distances showed similar correlations to assemblage dissimilarity between sites, being sometimes significantly correlated with biological variation of entire insect communities. In metacommunities deconstructed by body size or dispersal mode, contrary to our expectation, passive dispersers showed a slightly stronger correlation than active dispersers to environmental differences between sites, although passive dispersers also showed a stronger correlation than active dispersers to physical distances between sites. The strength of correlation between environmental distance and biological dissimilarity also varied slightly among the body size classes. 4. After controlling for environmental differences between sites, cumulative cost distances were slightly better correlates of biological dissimilarities than overland or watercourse distances between sites. However, quantitative differences in correlation coefficients were small between different physical distances. 5. Although environmental differences typically override physical distances as determinants of the composition of stream insect assemblages, correlations between environmental distances and biological dissimilarities are typically rather weak. This undetermined variation may be attributable to dispersal processes, which may be captured using better proxies for the process. We suggest that further modifying the measurement of cost distances may be a fruitful avenue, especially if complemented by more direct natural history information on insect dispersal behaviour and distances travelled by them.