The paper considers the problem min{f(x,y): gi(x,y)•0, i=1,...,m,x∈X,y∈Y∈ where f and the gi are lower semicontinuous and convex in y for fixed x but not convex jointly in (x,y), X is a compact subset of Rn,Y is a closed convex set in Rm. In order to decompose this problem into subproblems, each depending either on the x-variables alone or the y-variables alone, a new partitioning method is proposed which does not require the Benders-Geoffrion’s condition on the structure of joint constraints and the objective function. The relaxed master problems generated by the present partitioning method are d.c. programs and they can be systematically solved by recent algorithms.