A modeling system that couples a land-usebased export coefficient model, a stream nutrient transport equation, and Bayesian statistics was developed for stream nitrogen source apportionment. It divides a watershed into several sub-catchments, and then considers the major landuse categories as stream nitrogen sources in each subcatchment. The runoff depth and stream water depth are considered as the major factors influencing delivery of nitrogen from land to downstream stream node within each sub-catchment. The nitrogen sources and delivery processes are lumped into several constant parameters that were calibrated using Bayesian statistics from commonly available stream monitoring and land-use datasets. This modeling system was successfully applied to total nitrogen (TN) pollution control scheme development for the ChangLe River watershed containing six sub-catchments and four land-use categories. The temporal (across months and years) and spatial (across sub-catchments and land-use categories) variability of nonpoint source (NPS) TN export to stream channels and delivery to the watershed outlet were assessed. After adjustment for in-stream TNretention, the time periods and watershed areas with disproportionately high-TN contributions to the stream were identified. Aimed at a target stream TN level of 2 mg L-1, a quantitative TN pollution control scheme was further developed to determine which sub-catchments, which land-use categories in a sub-catchment, which time periods, and how large of NPS TN export reduction were required. This modeling system provides a powerful tool for stream nitrogen source apportionment and pollution control scheme development at the watershed scale and has only limited data requirements.