Trains are now recognized as powerful sources for seismic interferometry based on noise correlation, but the optimal use of these signals still requires a better understanding of their source mechanisms. Here, we present a simple approach for modeling train-generated signals inspired by early work in the engineering community, assuming that seismic waves are emitted by sleepers regularly spaced along the railway and excited by passing train wheels. Our modeling reproduces well seismological observations of tremor-like emergent signals and of their harmonic spectra. We illustrate how these spectra are modulated by wheel spacing, and how their high-frequency content is controlled by the distribution of axle loads over the rail, which mainly depends on ground stiffness beneath the railway. This is summarized as a simple rule of thumb that predicts the frequency bands in which most of train-radiated energy is expected, as a function of train speed and of axle distance within bogies. Furthermore, we identify two end-member mechanisms—single stationary source versus single moving load—that explain two types of documented observations, characterized by different spectral signatures related to train speed and either wagon length or sleeper spacing. In view of using train-generated signals for seismic applications, an important conclusion is that the frequency content of the signals is dominated by high-frequency harmonics and not by fundamental modes of vibrations. Consequently, most train traffic worldwide is expected to generate signals with a significant high-frequency content, in particular in the case of trains traveling at variable speeds that produce truly broadband signals. Proposing a framework for predicting train-generated seismic wavefields over meters to kilometers distance from railways, this work paves the way for high-resolution passive seismic imaging and monitoring at different scales with applications to near-surface surveys (aquifers, civil engineering), natural resources exploration, and natural hazard studies (landslides, earthquakes, and volcanoes).