During periods of reduced precipitation, flow in low-order, upland streams may be reduced and may stop completely. Under these "low flow" conditions, fecal bacteria directly deposited in the stream dominate in-stream bacteria loads. When developing a Total Maximum Daily Load (TMDL) to address a bacterial impairment in an upland, rural watershed, direct deposit (DD) fecal bacteria sources (livestock and wildlife defecating directly in the stream) often drive the source-load reductions required to meet water quality criteria. Due to limitations in the application of existing watershed-scale water quality models, under low-flow conditions the models can predict unrealistically high in-stream fecal bacteria concentrations. These unrealistically high simulated concentrations result in TMDL bacteria source reductions that are much more severe than what actually may be needed to meet applicable water quality criteria. This study used the Hydrological Simulation Program-FORTRAN (HSPF) to compare three low-flow DD simulation approaches and combinations (treatments) on two Virginia watersheds where bacterial impairment TMDLs had been previously developed and where low-flow conditions had been encountered. The three methods; Flow Stagnation (FS), DD Stage Cut-off (SC), and Stream Reach Surface Area (SA), have all been used previously to develop TMDLs. A modified version of the Climate Generation (CLIGEN) program was used to stochastically generate climate inputs for multiple model simulations. Violations of Virginia's interim fecal coliform criteria and the maximum simulated in-stream fecal coliform concentration were used to compare each treatment using ANOVA and Kruskal Wallis rank sum procedures. Livestock DD bacteria sources were incrementally reduced (100%, 50%, 15%, 10%, 5%) to represent TMDL load reduction allocation scenarios (allocation levels). Results from the first watershed indicate that the FS method simulated significantly lower instantaneous criterion violation rates at all allocation levels than the Control. The SC method reduced the livestock DD load compared to the Control, but produced significantly lower instantaneous criterion violation rates only at the 100% allocation level. The SA method did not produce significantly different instantaneous criterion violation rates compared to the Control. Geometric mean criterion violation rates were not significantly different from the Control at any allocation level. The distributions of maximum in-stream fecal coliform concentrations simulated by the combinations SC + FS and SC + SA + FS were both significantly different from the Control at the 100% allocation level. The second watershed did not produce low-flow conditions sufficient to engage the FS or SC methods. However, the SA method produced significantly different instantaneous violation rates than the Control at all allocation levels, which suggests that the SA method continues to affect livestock DD loads when low-flow conditions are not simulated in the watershed. No significant differences were found in the geometric mean violation rate or distribution of maximum simulated in-stream fecal coliform concentrations compared to the Control at any allocation level. This research suggests that a combination of the SC and FS methods may be the most appropriate treatment for addressing unrealistically high concentrations simulated during low-flow conditions. However, this combination must be used with caution as the FS method may increase the maximum simulated in-stream fecal coliform concentration if HSPF simulates zero volume within the reach. / Master of Science