Recently, the necessity of developing a practical bridge management system (BMS) has been pointed out in Japan, because the maintenance of existing bridges has become a major social concern. The aim of this study was to develop a practical BMS for deteriorated concrete bridges. The proposed system (J-BMS) uses multilayered neural networks to predict deterioration processes in existing bridges, to construct an optimal maintenance plan for repair or strengthening measures based on minimizing life-cycle cost, and to estimate the maintenance cost. A comparison of the results of applying this system to some actual in-service bridges with the results of questionnaire surveys of experts indicates that optimal maintenance planning as well as bridge rating can be predicted accurately by this system.