Due to the complexity of the production process, short-term scheduling for refineries is one of the most challenging problems. To address large-scale industrial refinery scheduling problems, a spatial Lagrangian decomposition approach is proposed to decompose the whole problem into several production processing subproblems and one blending and delivery subproblem. Some auxiliary constraints are added in the subproblems to accelerate the convergence of Lagrange multipliers. An initialization scheme of Lagrange multipliers, a hybrid method to update the Lagrange multipliers, and a heuristic algorithm to find feasible solutions are also designed. Computational results on three cases with different lengths of time horizons and different numbers of orders show that the proposed Lagrangian scheme is effective and efficient.