Reaches extending for tens or hundreds of channel widths along a river are normally seen as stable and homogeneous from a regime perspective, but other approaches emphasise channel change and within-reach spatial variability in flow and physical habitat. Tensions between different perspectives are discussed, with particular emphasis on limitations of traditional regime approaches. Cross-cutting issues include reconciling different scales of self-organisation in gravel-bed rivers, the need to treat bed characteristics as a degree of freedom, and within-reach spatial variability and temporal fluctuation. New ways to tackle these and other issues have been enabled by ever-increasing computing power. Non-uniform and/or unsteady fluvial processes can be modelled numerically, and remote-sensing methods have been developed to acquire dense spatially distributed measurements. But neither models nor observations are infallible, and models of different complexity need to be compared and assessed carefully.