In this paper, an autonomous decentralized scheduling system for just-in-time production is proposed. In the proposed system, each scheduling sub-system belonging to respective production stage derives a near optimal schedule by repeating the generation of the schedule of its own stage and data exchange among the other production stages. The objective function for each scheduling sub-system includes the storage costs for intermediate and final products in addition to the changeover costs and the due date penalties. When the storage costs are included in the objective function, the calculation of starting times of jobs becomes difficult even if the production sequence is given. In this study, an efficient algorithm is introduced to each scheduling sub-system to reduce the computational effort. It is shown that the performance of the proposed system is almost as good as that of a conventional scheduling system considering the entire plant. To cope with various unexpected events for the proposed system, a rescheduling procedure is developed and tested on a discrete-event simulator. The results show, if the proposed procedure is adopted, improved schedules, which minimize the effects of disturbances, can be generated in a significantly shorter computation time.