In the last two decades, numerous models and modeling techniques have been developed to simulate nonpoint source pollution effects. Most models simulate the hydrological, chemical, and physical processes involved in the entrainment and transport of sediment, nutrients, and pesticides. Very often these models require a distributed modeling approach and are limited in scope by the requirement of homogeneity and by the need to manipulate extensive data sets. Physically based models are extensively used in this field as a decision support for managing the nonpoint source emissions. A common characteristic of this type of model is a demanding input of several state variables that makes the calibration and effort-costing in implementing any simulation scenario more difficult. In this study the USDA Soil and Water Assessment Tool (SWAT) was used to model the Venice Lagoon Watershed (VLW), Northern Italy. A Multi-Layer Perceptron (MLP) network was trained on SWAT simulations and used as a meta-model for scenario analysis. The MLP meta-model was successfully trained and showed an overall accuracy higher than 70% both on the training and on the evaluation set, allowing a significant simplification in conducting scenario analysis.