Sediment routing models which simulate the coevolution of river long profile and bed grain size distributions have been used to investigate downstream fining above base level, channel response to rectification, and other disequilibrium situations at a variety of timescales. We extend one such model (SEDROUT <Hoey and Ferguson, 1994>) to deal with sand as well as gravel and test its ability to simulate aggradation and downstream fining in a well-documented Canadian river. Previous tests of this and similar models have been largely restricted to comparing observed and simulated gradients of downstream fining, but it is not obvious when to make the comparison in a time-dependent model with uncertain initial state and no equilibrium except in the very long term. We discuss and apply a more rigorous set of test criteria and address issues of defining initial conditions and time to test. The model's varying sensitivity to different boundary conditions and parameters indicates key data constraints on the testability and predictive accuracy of any such model. We also consider the adequacy of one-dimensional calculations in channels with variable width and present initial results of attempts to allow for this. ¿ 2001 American Geophysical Union