The routing of bed material through channels is poorly understood. We approach the problem by observing and modeling the fate of a low-amplitude sediment wave of poorly sorted sand that we introduced into an experimental channel transporting sediment identical to that of the introduced wave. The wave essentially dispersed upstream and downstream without translation, although there was inconclusive evidence of translation late in the experiment when the wave was only 10--20 grain diameters high. Alternate bars migrated through zones of differing bed load transport rate without varying systematically in volume, celerity, or transport rate. Sediment that overpassed migrating bars was apparently responsible for dispersion of the wave. The evolution of the wave was well predicted by a one-dimensional model that contains no adjusted empirical constants. Numerical experiments demonstrate, however, that the theory does not predict sediment waves that migrate long distances downstream. Such waves can only be explained by the following processes not represented by the theory: selective bed load transport, spatial variations in bar and other form roughness, the mechanics of mobile armor, and perhaps other mechanisms.¿ 1997 American Geophysical Union