While railway transport appears well suited to revenue management (RM), establishing it in practice appears difficult. To explain this, we investigate the long-term consequences of repeated transactions and reference pricing. We consider the implications of reference pricing based on an agent-based simulation of passenger railway RM. The model is empirically calibrated using data provided by a European long-distance railway operator. On the long term, reducing fares to induce additional demand can foil revenue gains when customers learn and communicate reference prices. Accordingly, knowing customers' tendency to build reference prices becomes crucial.