The problem of finding optimal inventory control schemes has received considerable attention in the literature for the case of one inventory at a given location in a production process or distribution system. Almost no analytical work has been devoted to systems of interacting inventories at different locations, and to the problem of selecting optimal inventory locations from a number of possible ones. This is the subject of the present paper. The traditional cost-of-shortage concept is given up in favor of a functional relation between expected demand and average delivery time to the customer. The measure of system performance is sales revenue minus inventory carrying cost. A method employing a combination of dynamic programming and a one-dimensional maximization procedure, yields the optimal inventory locations as well as the corresponding order rules.