Researchers are assessing the beneficial effects of conservation practices on water quality with hydrologic models. The assessments depend heavily on accurate simulation of water yield. This study was conducted to improve Soil and Water Assessment Tool (SWAT) hydrologic model daily water yield estimates in the Little River Experimental Watershed (LREW) in South Georgia. The SWAT code was altered to recognize a difference in curve number between growing and dormant seasons, to use an initial abstraction (la), of 0.05 S rather than 0.2 S, and to adjust curve number based upon the level of soil saturation in low-lying riparian zones. Refinements were made to two SWAT input parameters, SURLAG and ALPHA-BF, from a previous set of calibration parameters. The combined changes improved the daily Nash-Sutcliffe model efficiency (NSE) from 0.42 to 0.66 for water yield at the outlet of the 16.9 km2 subwatershed K of the LREW for the ten-year period 1995 - 2004. Further calibration of the SURLAG coefficient yielded the largest improvement of five alterations and changing la effected the next largest improvement. Over the 10-year investigation period, the model predicted annual average water yield within 1% of measured streamflow and deviation between observed and simulated values for stormflow was < 2.2%. Annual daily NSEs for each of the ten years were improved; for two years affected by seasonal tropical storm events, NSEs were changed from negative to positive values. The results of this study support the adjustment of the la ratio in the runoff curve number and suggest that additional changes to SWAT would improve water yield prediction for southern Coastal Plain locations.