Train timetables with a combination of short-length and full-length services can adapt to spatial and temporal variations in demand. However, with such timetables, some passengers may choose to catch an earlier short-length service and then transfer to a full-length service to reach their destinations rather than waiting for a crowded full-length service. This uncertain service choice behavior frequently occurs, which, however, was not well considered in most studies on demand-oriented timetabling. Therefore, this study presents a robust timetabling approach based on the scenario-based method considering uncertain passenger volumes and service choice behavior for choosing different types of train services. A customized decomposition-based method with an iterative solution procedure is designed to solve the proposed model. In each iteration, a multi-agent-based simulation algorithm is developed to update passengers’ travel utility and the service choice preference proportion based on the previous iteration's timetabling results. To obtain high-quality solutions in an acceptable computing time, a hybrid “ALNS + GUROBI” algorithm is developed to handle the timetabling problems for large-scale cases. The proposed method optimizes the train timetables for Xi'an Metro Line 3 in China. Our results indicate that the proposed method can account for uncertain passenger volumes and service choice behavior by adjusting the short-length plans and train headways to match the transport capacity to passenger demand.