Bed load sediment was caught in pit traps at several locations on a bar in Harris Creek, a cobble-gravel stream with a nivally dominated hydrograph, a structurally highly organized bed, and very low rates of bed material transport. Observations were made during two spring freshets. In order to obtain representative grain size distributions of the material in transport, the traps were left for periods of up to 24 hours, so that samples of up to 30 kg were recovered. We examine the sensitivity of bedload flux to flow variations via trap-specific ratings for narrowly defined textural subranges. All the ratings are very sensitive, indicating that bed load flux remains in the regime of partial transport. The ratings also exhibited seasonal hysteresis and varied from trap to trap and from year to year. At one trap the ratings for large material are distinctly segmented, with no strong correlation for the highest flows. Significant transport for material >8 mm in size begins at ~7 m3 s-1, considerably higher than for sand. At flows competent to move the local bed gravel the portion of the sand load between 0.25 and 0.50 mm goes into suspension (finer material being dominantly suspended at all flows). Most of the sand transported at <7 m3 s-1 is washed in from upstream while the local bed remains stable. Auxiliary tracer studies demonstrate that at no observed stage (up to mean annual flood level) was the local bed generally mobilized. ¿ 2001 American Geophysical Union