In many arid and semi-arid regions whether or not to desalinate seawater has long been a non-issue and policy debates are focused on the timing and extent of the desalination activities. We analyze how water scarcity and demand structure, on the one hand, and cost reduction via R&D programs, on the other hand, affect the desirable development of desalination technologies and the time profiles of fresh and desalinated water supplies. We show that the optimal R&D policy is of a non-standard most rapid approach path (NSMRAP) type, under which the state of desalination technology - the accumulated learning from R&D efforts - should approach a pre-specified target process as rapidly as possible and proceed along it thereafter. The NSMRAP property enables a complete characterization of the optimal water policy. The renewable nature of the fresh water stock permits a non-monotonic behavior of the optimal stock process: under certain conditions, the stock is depleted, to be (fully or partly) refilled at a later date. © 2000 Elsevier Science B.V. All rights reserved.