The paper describes an efficient algorithm, based on a 1st-order gradient technique in conjunction with a nonlinear-programming technique, for the long-term scheduling of multistorage hydroelectric and multithermal systems for the minimum expected fuel cost under the constraints of expected water available for hydroelectric generation in a given period of time. The system variables are in a discrete form, and the water inflows and load demand are stochastic, their probability properties being pre-estimated from past history. It has been demonstrated through an example solved on an IBM 1130 digital computer that considerable saving in the cost can be achieved through the optimal allocation of the available water.