Abstract Bed load transport rates are difficult to predict in channels with bed material composed of sand and gravel mixtures. The transport of bed load was measured on Goodwin Creek, and in a laboratory flume channel with a similar bed material size distribution. The range of bed load transport rates measured in the laboratory channel were similar to those measured in the channel of Goodwin Creek; however, the shear stresses calculated on Goodwin Creek were three times greater than in the laboratory channel for similar bed load transport rates. Much of this difference in shear stress was removed by applying the drag partitioning technique of Einstein, although significant differences between the two sets of data remain. Predictions of bed load transport rates using three previously published transport relations were good for most flows for the laboratory data. For the Goodwin Creek bed load data, predicted transport rates were close to measured ones for low flows but diverged by an order of magnitude or more for high shear stresses. Improved methods of shear stress partitioning are needed to improve the performance of bed load transport relations on streams of this type.