This study is to check the applicability of SWAT-APEX (Soil and Water Assessment Tool-Agricultural Policy / Environmental eXtender) model as combined watershed and field models by applying the APEX to paddies in a watershed (465.1 $km^2$) including Yedang reservoir. Firstly, the SWAT were calibrated with 3 years (2000~2002) daily streamflow and monthly water quality (T-N and T-P) data, and validated for another 3 years (2003~2005) data. The average Nash-Sutcliffe model efficiency (ME) of streamflow during validation was 0.73, and the coefficient of determination ($R^2$) of T-N and T-P were 0.77 and 0.73 respectively. Next, running the SWAT-APEX model with the SWAT calibrated parameters for paddies, the $R^2$ of T-N and T-P were 0.80 and 0.76 respectively. The results showed that SWAT-APEX model was more correctly predicted for T-N and T-P loads than SWAT model. The difference results between watershed and field models was predicted to have substantial impact on NPS loads, especially on T-N and T-P loads. Therefore, to improve negative NPS load simulations should be considered the model characteristics as simulating mechanism to properly select the NPS model for agricultural watershed.