Many rail terminals have loading areas that are properly equipped to move containers between trains. With the growing throughput of these terminals all the trains involved in a sequence of such movements may not ¿t in the loading area simultaneously, and storage areas are needed to place containers waiting for their destination train, although this storage increases the cost of the transshipment. This increases the complexity of the planning decisions concerning these activities, since now trains need to be packed in groups that ¿t in the loading area, in such a way that the number of containers moved to the storage area is minimized. Additionally, each train is only allowed to enter the loading area once. Similarly to previous authors, we model this situation as an acyclic graph partitioning problem for which we present a new formulation, and several valid inequalities based on its theoretical properties. Our computational experiments show that the new formulation outperforms the previously existing ones, providing results that improve even on the best exact algorithm designed so far for this problem.