The interface system, GIS-AGNPS was to be validated with field data from six tested small watersheds ranging from 0.7 to 4.7$km^2$ in size which have steep topography and complex landuses. The model validation involved the calibration of input parameters and component modifications, in efforts to develop a model applicable to general uses for identifying and controlling nonpoint source pollution loads from agricultural watersheds. The simulated direct runoff from AGNPS was in good agreement with the field data for the averaged antecedent moisture conditions or AMC- II. The results differed, however, from the observed for AMC- I or III. A simple empirical relationship was proposed to estimate the curve number for AMC- I or m from AMC- II, which was found to result in simulated runoff close to the observed. The peak runoff relationship at AGNPS was also modified to reflect the watershed conditions and tested satisfactorily with the field data. The simulated sediment yields from the watersheds were fair as compared to the observed. Nutrient loads simulated from the model were different from the observed data. It appeared that the model was incapable of adequate depicting nutrient transport processes at paddy field and other landuses of the tested watersheds. Some modifications may be needed for the accurate representing the processes at paddy field.