The Japanese Bridge Management System (J-BMS) RC version consists of the following three subsystems: (1) a bridge maintenance database system (J-BMS DB), (2) a concrete bridge rating expert system (RC-BREX) and (3) a maintenance plan optimization system (MPOS). Especially, in this paper, the RC-BREX system for deteriorating concrete bridges is described based on how to construct from a hierarchical neural network in order to carry out fuzzy inference and machine learning by using the neural network with the back-propagation method. Furthermore, the comparisons between diagnostic results by bridge experts and those of the proposed system are presented so as to demonstrate the validity of the system’s learning capability by using the training set for machine learning which obtained from inspection on actual in-service bridges and questionnaire surveys by bridge experts.