This paper investigates the feasibility of predicting nitrate contamination from agricultural sources using Multi-Layer Perceptron Artificial Neural Networks (MLP-ANNs). The approach consists in training an MLP-ANN to predict nitrate concentrations based on a set of indirect measurements, such as pH, electrical conductivity, temperature and groundwater level. These are simpler and more economical than direct measurements, and they can be continuously collected on-site, rather than by performing laboratory tests. The approach has been validated in the nitrate vulnerable zone of the Arborea plain (central western Sardinia, Italy) by comparing the results obtained with different MLP-ANN models in order to find the most efficient model. The results show that the MLP-ANN-based model is a time- and cost-efficient method for predicting nitrate concentration.