The Lagrangean relaxation method is applied to the unit commitment problem of a power generation system with pumped-storage units. The shortest path algorithm is used for the subproblem for each thermal unit and the minimal cost flow algorithm is used for each pumped-storage unit. A method for finding a better feasible solution from a solution of a relaxed problem is presented. A real Korean power generation system with 39 thermal units and one pumped-storage unit is tested. We considered 168 planning hours and we were able to find a near optimal solution (within 0.5% of optimal) consuming a reasonable amount of CPU time.