This paper presents an approach for allocating spares on three echelons to increase system reliability and decrease plant down-time in a cost-effective manner. The usual assumptions are made: a) the demands for spares have the Poisson distribution with known demand rates, b) all the failures of plant parts are s-independent, c) for the plant to be operating, all parts must be operating, thus forming 1-out-of-M:F configuration. To each echelon of the spatially distributed spare-support system is associated a relevant optimization criterion and a corresponding constraint. The algorithm for determining stock levels adapts the known `per-pound' procedure, applied in sequence from the first echelon to the third echelon. The sequential optimization is computationally efficient. The test examples have confirmed that the algorithms proposed can handle 3-echelon sparing problems with even thousands of part types.