The ecosystem concept should be reappraised as a basic model for rivers, with regard for flow as an organizing variable. This would facilitate comparisons between the large rivers of humid climates, where flow regimes are comparatively regular, and those of arid and semi-arid areas, where river regimes are highly variable. Ecosystem processes might be modelled by combining the river continuum and flood pulse concepts, with refinements to accommodate a complex flood pulse (e.g. variations in stage amplitude, timing, duration, rates of rise and fall). Patch boundaries (ecotones) such as the riverine littoral zone warrant close study because they strongly influence the structure and dynamics of the ecosystem. The general model needs a quantitative basis, perhaps focused on the balance of processes involved in the physical transport and biological transformation of carbon. The ultimate test of such a model will be in its capacity to predict the effects of flow regulation. Further development, however, is limited by data. In both research and management monitoring programmes need to be established to provide information and to develop a sustained, comprehensive approach to dryland rivers as ecosystems.