This study tested and evaluated the agricultural non-point source (AGNPS) model for the Wuchuan catchment, a typical agricultural area in the Jiulong River watershed, Fujian Province, China. The AGNPS model was calibrated and validated for the study area with observed data on ten storms. The data on eight storms in 2002 were used for calibration while data on two storms were used for validation of the model. Considering the lack of water quality data over a long-term series, a novel method, comparing an internal nested catchment with its surrounding catchment, was used to supplement the less long-term series data. Dual calibration and validation of the AGNPS model was obtained by this comparison. The results indicate that the correlation coefficients were 0.99 and 0.98 for runoff, 0.94 and 0.95 for the peak runoff rate of the large catchment and the small catchment, respectively, and 0.76 for the sediment of the small catchment only. Each pair of correlation coefficients is homogeneous for the same event for the two catchments. With the exception of the sediment yield and particulate phosphorus, the peak runoff rate and other nutrients were well predicted. Sensitivity analysis showed that the Soil Conservation Service curve number and rainfall quantity were the most sensitive parameters, which resulted in high output variations. Erosivity and other parameters had little influence on the hydrological and quality outputs.