This document summarizes a study that assessed the performance of four artificial neural network (ANN) models for predicting groundwater levels. The four ANN models used different transfer functions in their hidden layers: binary sigmoid, logistic sigmoid, hyperbolic tangent, and Zakharov-Lebedev bounded sigmoid. The ANNs were trained on synthetic groundwater level data from a numerical model and then their predictions were evaluated using both statistical and graphical techniques. The statistical analysis showed that the ANN using the hyperbolic tangent transfer function had the best performance based on the lowest errors, biases, and index values. The graphical techniques also indicated this ANN provided the most accurate nonlinear predictions of the groundwater levels.