Technical Abstract: Surface runoff and sediment transport influence soil quality and the quality of receiving waters. Simulation models are useful tools to describe drainage behavior and sediment yield to develop management strategies for agricultural waters. The objective of this study was to investigate the performance of the Agricultural Non-Point Source Pollution Model (AGNPS) and modified versions comparing measured and predicted surface runoff and sediment yield at the drainage outlet. The study was carried out using 52 rainfall-runoff events from two small watersheds in Bavaria, 22 for calibration and 30 for validation. Evaluation of model outputs was based on graphical displays contrasting measured and predicted values for each rainfall-runoff event, and standard statistics such as coefficient of efficiency. A comparison between 3 different surface runoff methods Q1 (uncalibrated curve number (CN) method), Q2 (calibrated CN method), and Q3 (Lutz method) showed that the uncalibrated CN method underestimated measured surface runoff considerable, while the Lutz method outperformed the calibrated CN method and was better than the uncalibrated CN method. The modifications made to sediment yield calculations encompassed: (i) replacement of the Universal Soil Loss Equation LS factor algorithm by one based on stream power theory (variant S2), and (ii) linkage of channel erosion by individual categories of particle size to runoff velocity (variant S3). The sediment yield predictions by S2 and S3 outperformed the predictions of S1 (USLE). Calculations based on S3 gave the best agreement when compared to measured sediment yield values at the drainage outlet.