Abstract: China-Europe freight train is an international container rail freight train between China and Europe and countries along the Belt and Road.Since its inception, the two sides have carried out fruitful cooperations under the framework of "Belt and Road Initiative".From the current development situation, the freight service will usher in a broader prospect, so the preparation of the operation scheme is particularly important.The preparation of the operation scheme usually takes the economic benefits of the railway operator, and rarely considers the needs both of operatorand customer.An optimization method for operation schemes of China-Europe Railway Express based on the demand chain is proposed, taking into account factors from both the railway operator and customer perspectives.The formulation of operation schemes is achieved in two stages.Firstly, a candidate set is generated based onthe shortest path principle, resulting in an initial set of solutions for China-Europe Railway Express operation schemes, effectively increasing the computational complexity of the model.Secondly, a bi-level programming model for China-Europe Railway Express operation schemes is constructed based on the "one arc, more services, more flow" network.The upper-level model, from the perspective of the railway operator, considers constraints such as transportation time, carbon emissions, and capacity limitations, aiming to maximize transportation benefits.The lower-level model, based on actual customer demand, extends to a logistics service selection model that allows for flow splitting, and a particle swarm algorithm embedded with the Lagrangian relaxation method is designed for solving the model.Through the interaction between the upper and lower-level models, the game behavior of supply and demand is demonstrated in the market, thus applying the concept of demand chain management to China-Europe Railway Express operation schemes.The example shows that when the freight flow allows splitting, the railway operator’s profit increases by 41.67%,which fully confirms the superiority of this model.