Uncertainty of best management practice (BMP) effectiveness is an important factor in the development of watershed management plans. This study explored the uncertainty of BMP effectiveness in reducing total nitrogen (TN) load owing to the uncertainty in hydrological parameters, thereby improving their reliability. A watershed model, annualized agricultural non-point source pollution (AnnAGNPS), was employed to evaluate the effectiveness of the four potentially feasible BMPs (i.e., riparian buffer, fertilization reduction, no-tillage, and parallel terraces) in the Shanmei Reservoir watershed, located in the southeastern coastal region of China. Annual and seasonal uncertainty variations in BMP effectiveness were evaluated based on ten parameter sets selected from 1000 parameter groups using Latin hypercube sampling. The results showed that the uncertainty of BMP effectiveness in reducing the TN load was larger than the uncertainty of TN load simulation at annual and seasonal time scales. The BMP effectiveness tended to be higher in summer than in the other seasons. The uncertainty of BMP effectiveness varied seasonally, and it was always lower in summer for most BMPs. This indicated that the impact of BMPs on reducing TN load was more effective, with a higher reduction rate and lower uncertainty in summer. Among the BMPs, the parallel terrace was the most effective measure for reducing TN load since it had the highest reduction rate and relatively low uncertainty. Although this study is a case study, it can provide a scientific reference for decision-making in uncertain situations when AnnAGNPS is applied for water quality simulations.