This study presents the design and experimental testing of a scaled-down steel railway bridge model for testing vibration-based damage identification algorithms. This model represents a 2 m-long, three-span continuous bridge, crossed by a remotely controlled tandem vehicle simulating train load effects. Particular attention has been paid to properly define the boundary conditions: fixed at one abutment and roller-pin supports at intermediate piers and the second abutment. Custom support devices with arrays of linear springs have been designed to control both vertical and torsional stiffness. The model and the vehicle are densely instrumented using low-cost MEMS accelerometers. The experimental setup enables testing across a wide range of operational conditions (i.e., train speed and mass) and damage scenarios (i.e., losses of vertical and torsional stiffness of the supports, and deck-localized damage simulated by point masses). This work reports the generation of an extensive monitoring database, along with the finite element modelling and inverse calibration of the physical model. A detailed analysis and discussion is provided on the impacts of different damage scenarios on both the modal properties and the transient dynamic response under train loading. The resulting database supports the evaluation of various damage identification algorithms, promoting their future implementation in real-world settings.