We develop a new method for analysis of meandering channels based on planform sinuosity. This analysis objectively identifies three channel-reach lengths based on sinuosity measured at those lengths: the length of typical, simple bends; the length of long, often compound bends; and the length of several bends in sequence that often evolve from compound bends to form multibend loops. These lengths, when normalized by channel width, tend to fall into distinct and clustered ranges for different natural channels. Mean sinuosity at these lengths also falls into distinct ranges. That range is largest for the third and greatest length, indicating that, for some streams, multibend loops are important for planform sinuosity, whereas for other streams, multibend loops are less important. The role of multibend loops is seldom addressed in the literature, and they are not well predicted by previous modelling efforts. Also neglected by previous modelling efforts is bank–flow interaction and its role in meander evolution. We introduce a simple river meandering model based on topographic steering that has more in common with cellular approaches to channel braiding and landscape evolution modelling than to rigorous, physics-based analyses of river meandering. The model is sufficient to produce reasonable meandering channel evolution and predicts compound bend and multibend loop formation similar to that observed in nature, in both mechanism and importance for planform sinuosity. In the model, the tendency to form compound bends is sensitive to the relative magnitudes of two lengths governing meander evolution: (i) the distance between the bend cross-over and the zone of maximum bank shear stress, and (ii) the bank shear stress dissipation length related to bank roughness. In our simple model, the two lengths are independent. This sensitivity implies that the tendency for natural channels to form compound bends may be greater when the banks are smoother. Copyright © 2002 John Wiley & Sons, Ltd.