Along the San Bernardino Valley, the Santa Ana River decreases in slope, increases in width, and deposits particles from boulders to sand as it loses transport capacity. Episodic rainfalls feed very large winter floods, but dry summer and fall periods lead to extensive dry alluvial reaches due to surface water infiltration into subsurface aquifers. Within one of these dry reaches, a small inset channel has developed to effectively convey year-round wastewater discharges. This flow creates a coarse bed substrate composed of gravel and cobble that is home for diatoms and algae, the diet for the Santa Ana Sucker, a threatened native fish. Field-based observations include: • Shear stresses in the inset channel are capable of transporting sand as bedload, but not gravel and cobble. Bedload measurements (at several locations and times under wastewater discharge conditions) indicate that about 90% of the bedload is sand. The median grain size of bedload samples is consistently in the 0.5–1-mm range with little longitudinal variability. • The upstream supply of sand decreases through time as winnowing proceeds in the downstream direction. Bedload transport rates increase in the downstream direction by an order of magnitude over just 7 km, despite the fact that water discharge decreases downstream due to infiltration (typically by 25–50% over the 7-km reach). • The location of the gravel-to-sand bed transition is variable from year to year and is related to the amount of time that has elapsed since the most recent upstream runoff event. These events supply new sediment to the reach and reset the winnowing process. • Direct observations of the inset channel following upstream runoff events revealed many instances of large areas of substantial sand deposition in the channel, in areas that were previously entirely gravel and cobble sizes. We propose a qualitative function to predict the bed substrate of the inset channel at any given time and location. A one-dimensional model of bedload transport and bed substrate in the study reach is being developed to quantify and test the predictive function. Preliminary results indicate that the winnowing process is quite sensitive to wastewater discharge levels. The calibrated model will be used to assess how treatment plant discharge scenarios might be used to strategically manage winnowing rates and the amount of gravel and cobble exposed on the bed in the reach.