Dryland ecosystems are particularly vulnerable to erosion generated by livestock grazing. Quantifying this risk across a variety of landscape settings is essential for successful adaptive management, particularly in light of a changing climate. In the Upper Colorado River Basin, there are nearly 25 000 km² of rangelands with underlying soils derived from Mancos Shale, an erodible and saline geologic parent material. Salinity is a major concern within the Colorado River watershed, much of which is attributed to runoff and leaching from Mancos Shale deposits. In a 60-yr paired-watershed experiment in western Colorado, we used silt fences to measure differences in saline hillslope erosion, including both total sediment yield and concentrations of primary saline constituents (Na and Se), in watersheds that were either exposed to grazing or where livestock was excluded. After accounting for the strong effects of soil type, slope, and antecedent precipitation, we found that grazing increased sediment loss by ≈50% across our 8-yr time series (0.1–1.5 tn ha^–1), consistent with levels reported at the watershed scale in early published work from studies at the same location. Eroded sediment Se levels were low and unaffected by grazing history, but Na concentrations were significantly reduced on grazed hillslopes, likely due to depletion of surface Na in soils exposed to chronic soil disturbance by livestock. Variable selection and path analysis identified that biological soil crust (BSC) cover, more than any other variable, explained the differences in sediment yields between grazed and ungrazed watersheds, partially through the enhancement of soil aggregate stability. Our results suggest that BSC cover should be granted heightened consideration in rangeland decision support tools (e.g., state-and-transition models) and that measures to reduce surface disturbance from livestock such as altering the timing or intensity of grazing may be effective for reducing downstream impacts.