Efforts to develop practical expert systems have been mostly concentrated on how to implement experience-based machine learning successfully. Recently several active researches on machine learning have been undertaken from the viewpoints of knowledge base management. The aim of this study is to develop the Concrete Bridge Rating (Diagnosis) Expert System with machine learning employing the combination of neural networks and bidirectional associative memories (BAM). Introduction of machine learning into this system facilitates knowledge base refinement. By applying the system to an actual in-service bridge, it has been verified that the employed machine learning method using results of questionnaire surveys on bridge experts is effective for the system.