River networks are hierarchical dendritic habitats embedded within the terrestrial landscape, with varying connectivity between sites depending on their positions along the network. This physical organisation influences the dispersal of organisms, which ultimately affects metacommunity dynamics and biodiversity patterns. We provide a conceptual synthesis of the role of river networks in structuring metacommunities in relation to dispersal processes in riverine ecosystems. We explore where the river network best explains observed metacommunity structure compared to other measurements of physical connectivity. We mostly focus on invertebrates, but also consider other taxonomic groups, including microbes, fishes, plants, and amphibians. Synthesising studies that compared multiple spatial distance metrics, we found that the importance of the river network itself in explaining metacommunity patterns depended on a variety of factors, including dispersal mode (aquatic versus aerial versus terrestrial) and landscape type (arid versus mesic), as well as location-specific factors, such as network connectivity, land use, topographic heterogeneity, and biotic interactions. The river network appears to be less important for strong aerial dispersers and insects in arid systems than for other groups and biomes, but there is considerable variability. Borrowing from other literature, particularly landscape genetics, we developed a conceptual model that predicts that the explanatory power of the river network peaks in mesic systems for obligate aquatic dispersers. We propose directions of future avenues of research, including the use of manipulative field and laboratory experiments that test metacommunity theory in river networks. While field and laboratory experiments have their own benefits and drawbacks (e.g. reality, control, cost), both are powerful approaches for understanding the mechanisms structuring metacommunities, by teasing apart dispersal and niche-related factors. Finally, improving our knowledge of dispersal in river networks will benefit from expanding the breadth of cost-distance modelling to better infer dispersal from observational data; an improved understanding of life-history strategies rather than relying on independent traits; exploring individual-level variation in dispersal through detailed genetic studies; detailed studies on fine-scale environmental (e.g. daily hydrology) and organismal spatiotemporal variability; and synthesising comparative, experimental, and theoretical work. Expanding in these areas will help to push the current state of the science from a largely pattern-detection mode into a new phase of more mechanistically driven research.