The paper estimates the most important measures of performance, specifically, the long-run average backorders at the retailers, and the average inventory at each location. These estimates are used within a cost-minimization framework to choose among policies, though the estimates could be used in other ways also. This approach can be described as a decomposition technique: each facility is approximated as a single-location inventory system. However, the parameters describing these single-location systems are dependent on the policies and the performance of other facilities. The same problem was examined by Deuermeyer and Schwarz (1981). The authors are following here the approach pioneered by Deuermeyer and Schwarz, who first suggested decomposition as an adaptation of the METRIC technique for this problem. The key innovation in the models is the use of second-moment information in the approximations.