This study proposes an integrated inference system to predict the financial performance of banks. The model comprises of two stages. At the first stage, the dominance-based rough set approach (DRSA) method is applied to reduce the complexity of the attributes involved, and the obtained decision rules are further refined by the neuro-fuzzy inference technique to indicate the fuzzy intervals for each attribute. The proposed model not only shows how to explore the implicit patterns regarding the bank’s performance change, but also refines the knowledge by tuning the parameters of membership functions for each attribute. At the second stage, the directional influences among the core attributes are further explored. To examine the proposed model, a group of real commercial banks in Taiwan is analyzed to construct the model, and five sample banks are tested to validate its effectiveness. The result provides understandable insights regarding the performance prediction problem of banks.