An experiment of falling wheels was performed in this study along with full-field tests, for a type of subway train on four subway lines. The results confirmed that 40–80 Hz vibration signals on the ground and train are caused by the wheel–rail P2 force. Vehicle–track-coupled dynamics models were created where the single-layer elastic support model, double-layer elastic support model and floating slab track model were considered. The results were validated by available experimental data and three types of theoretical models: Jenkins, Winkler and double-beam model. This shows that the amplitude of the P2 force increases with larger suspended rail joint angle, higher train speed, heavy unsuspended mass and stronger fastener stiffness. Besides, both fastener damping and the vertical stiffness of primary suspension enabled control of the amplitude of the P2 force in a limited interval. Supporting stiffness of track and unsprung mass played a key role in the frequency of the P2 force. However, the vertical stiffness of the primary suspension had a reduced effect on the frequency of P2 force as compared with the supporting stiffness. Thus, for an effective change to the supporting stiffness, the fastener stiffness could be changed accordingly.