Due to limitations of both the limited size of the acoustic wind tunnel and the computational resources, the study of aerodynamic noise in high-speed trains typically employs scaled models. However, scaled models result in reduced Reynolds numbers, altering flow field structures and affecting radiated noise. To delve deeper into the effects of scaling on aerodynamic noise in the pantograph region, this study employed scaled models at ratios of 1/4, 1/8, 1/16, 1/25, and 1/50, using Large Eddy Simulation (LES) and the Ffowcs Williams-Hawkings (FW-H) integral equation. The aerodynamic disturbance characteristics, surface dipole sound energy, and far-field radiation sound pressure levels in the overhead pantograph area of high-speed trains were investigated for different scale models. The research found that the transition in the flow field leads to the migration of sound sources within the pantograph, subsequently influencing its radiated noise characteristics. Research findings reveal that as model sizes decrease, the primary flow scale in the pantograph region shifts from the upper part to the lower part. This transition in the flow field leads to the migration of sound sources within the pantograph, subsequently influencing its radiated noise characteristics. As the size of the pantograph model decreases, radiated noise decreases. The sound pressure levels of the scaled models at 1/4, 1/8, 1/16, 1/25, and 1/50 ratios are 97.40 dB(A), 95.39 dB(A), 94.30 dB(A), 92.48 dB(A), and 88.1 dB(A), respectively. In the 1/4 scale model, the upper pantograph source contributes to 32% of the total sound power, and the overall acoustic spectrum in the pantograph exhibits two peaks with peak frequencies generated by the pantograph head and upper arm. In the 1/50 scaled model, the lower pantograph source contributes to 87% of the total sound power, and the overall acoustic spectrum in the pantograph exhibits a broad frequency characteristic, consistent with the pantograph cavity. The research findings reveal that the errors resulting from model scaling-down are non-linear. Therefore, appropriate corrections should be made during scale-down experiments based on the scaling factor and different acoustic characteristics. © 2024, Central South University Press. All rights reserved.