Despite increasing numbers of agricultural best management practices (BMPs) on the landscape and billions of dollars spent on BMP programs, stream health continues to decline in Virginia due to agricultural nonpoint-source pollution. The efficacy of BMPs for restoring stream health is potentially enhanced or constrained by conditions in the stream and watershed, as well as by social factors that determine landowner participation in BMP programs. We seek to understand BMP efficacy using an interdisciplinary approach involving watershed modeling, examination of ecological relationships with BMP installation, and assessments of landowner behaviors and perspectives regarding BMP programs. We used the Soil and Water Assessment Tool (SWAT) to model sediment and nutrient yields to streams in the upper Clinch, Powell, and Holston river watersheds in Virginia. The SWAT model generally predicted streamflow well and revealed critical-source areas of sediment, phosphorus, and nitrogen. Then we applied the SWAT results to select watersheds representing gradients in pollutant loading and BMP implementation. Currently, we are collecting ecological data from representative streams in those watersheds to test for effects of BMPs on stream health. Future work will use mail surveys conducted in the same watersheds chosen for the assessment of stream health to determine factors that lead to landowner persistence in BMP programs after cost-share contracts end. Finally, we will integrate this social and ecological information and design scenarios that assess the relative benefits to stream health derived from increasing persistence, increasing adoption of BMPs, or changing the types of BMPs adopted. The results of our research will provide managers with maps of critical-source areas where BMP placement is most needed, recommendations for the most appropriate BMPs depending on the landscape conditions, and tools for effectively encouraging landowner persistence in BMP programs at the conclusion of cost-share programs.