A method is proposed for minimizing a function of the form q(x-d)+f1(x)+ëëë+fm(x) where q is definite quadratic and fi are proper closed convex. The key feature of the method is its capability of reducing the problem to a sequence of subproblems of the form: minq(x-z)+fi(x). The method is an extension of the successive projection method given in [1] and has some similar features as the partial inverse method of Spingarn. In combination with the proximal point algorithm, it can decompose a general convex program. It has attractive convergence properties and is useful for solving large-scale sparse problems.