Credit rating for commercial loans is an important task for loan officers of a bank who usually use a credit-rating table(CRT) based on a point system. However, the loan decision making using such a table has its limitations as follows. First, CRT needs a few critical judgments for commercial loans. For example, the judgments consist of knowledge acquisition on the criteria such as variable selection, weights between variables, cut-offs between intervals within each variable, and relative value at each interval of the variables for credit rating. The first level criteria such as quantitative and qualitative variables need to be expressed on interval scales to assess a credit score for each variable. However, the variables present a problem in terms of measurability. This problem renders it impractical to directly apply theoretical or human expertise-based choice strategies to unstructured choice problems.The objective of this study is to present an optimal knowledge acquisition and refinement method of the CRT using genetic algorithms(GAs). In our proposed model, the GAs search the optimal criteria for the CRT applied to credit rating process. For the purpose, the CRT and the GAs are integrated to extract and refine the knowledge of multi-criteria for the CRT. The obtained knowledge supports to find optimal parameters of the CRT for loan officers' decision making. Our experimental results show that the knowledge acquisition and refinement method using the GAs are effective for the CRT as a multi-attribute decision problem.