For track maintenance to be both effective and economical, it is essential to predict the dynamic behavior of rolling stock when subjected to track irregularities. For this purpose, we identified rolling stock lateral dynamic characteristics and created models to predict this dynamic behavior in vehicles on curved track sections. The actual noncompensated lateral acceleration was used as the input signal on the model in order to predict lateral acceleration at low frequencies, and alignment and cross-level for that in the middle frequency range. In the middle frequency range, the lateral axle force prediction model was the same as that used for lateral acceleration. With these models, it proved possible to understand the relationship between track geometry and rolling stock lateral behavior.