A decomposition technique for linear programs is presented, in which the master program distributes the common resources and aims directly among the subprograms, rather than using price setting as is done in the Dantzig-Wolfe method. The technique is essentially a dual formulation of the Dantzig-Wolfe method. Consequently the optimum is reached in a finite number of steps. This is in contrast with the Kornai-Liptak method.