To properly investigate a driver behavior at signalized intersections, an artificial neural network (ANN) architecture called back-propagation network (BFN) was proposed in this study. Driver behavior can be modeled as a binary decision which is stop or go at the onset of amber. Field data were recorded from three intersections in Taiwan (one in Taipei city, two in Taichung city). The data were analyzed through the factor analysis and logit model estimation techniques. The BPN model developed for each intersection consisted of two input variables in the input buffer and one driver's decision in the output layer. The results obtained from the study show the applicability and validation of the basic ANN method to the complex driver behavior problem. The relatively simple BPN model turns out to be a very satisfactory tool for predicting accurate network outcomes compared to the logit model. It is expected that the validated model can be directly applied to traffic controls in urban areas.