Railway noise is an increasing concern in densely populated urban corridors, where reliable attenuation models are essential for land-use planning. However, field-based comparisons of different train categories under consistent conditions remain limited, particularly in Iran. This study investigated noise propagation from passenger trains, freight trains, and railbuses along a 4.2-km urban section of the Tehran–Tabriz railway near Kavousieh Station, Qods City. Measurements were conducted at seven distances (7.5–200 m) during three pass-bys for each train type. Smartphone measurements were cross-calibrated with a reference sound-level meter, and, after background-noise correction, spatial noise distributions were mapped using Ordinary Kriging. Train-specific logarithmic attenuation models were then developed and validated using grouped leave-one-distance-out cross-validation. Passenger trains produced the highest noise levels, reaching about 99 dB(A). The models achieved R² values of 0.9700–0.9845, demonstrating excellent agreement between predicted and measured noise levels. Using a 60 dB(A) criterion, estimated setback distances were approximately 191, 160, and 103 m for passenger trains, freight trains, and railbuses, respectively. Compared with the normalized FRA benchmark, the proposed models showed substantially smaller prediction errors at 200 m. These findings provide practical attenuation relationships for urban railway planning and establish a baseline for future railway acoustics and noise prediction studies in Iran.