Abstract. Credit rating agencies are being widely criticized because the lack of transparency in their rating procedures and the huge impact of the ratings they disclose, mainly their sovereign credit ratings. However the rationale seems to be that although credit ratings have performed worse than their aim, they are still the best available solution to provide financial markets with the information that their participants base their decisions on. This research work proposes a neural network system that simulates the sovereign credit ratings provided by two of the most important international agencies. Results indicate that the proposed system, based on a three layers structure of feed-forward neural networks, can model the agencies' sovereign credit ratings with a high accuracy rate, using a reduced set of publicly available economic data. The proposed model can be further developed in order to extent the use of neural networks to model other ratings, create new ratings with specific purposes, or forecast future ratings of credit rating agencies.