In this paper we present a decomposition strategy for solving large scheduling problems using mathematical programming methods. Instead of formulating one huge and unsolvable MILP problem, we propose a decomposition scheme that generates smaller programs that can often be solved to global optimality. The original problem is split into subproblems in a natural way using the special features of steel making and avoiding the need for expressing the highly complex rules as explicit constraints. We present a small illustrative example problem, and several real-world problems to demonstrate the capabilities of the proposed strategy, and the fact that the solutions typically lie within 1-3% of the global optimum.