Abstract: In low-light and high-dust concentration environments of fully mechanized mining faces, ground-penetrating radar (GPR) is an effective method for addressing the challenge of coal-rock interface identification. To address the issue of existing research neglecting the impact of rough coal-rock interfaces on identification results, this study employs the Monte Carlo method to generate rough surfaces with varying root mean square (RMS) heights and correlation lengths. These surfaces are superimposed onto a coal-rock interface with undulating geological features. Using the finite-difference time-domain (FDTD) method, forward modeling experiments are conducted to investigate the impact of coal-rock interface roughness on the position of the interface and the amplitude of backscattered waves. The experimental results indicate that when the RMS height of the rough coal-rock interface exceeds 1/4 of the wavelength, multipath effects cause electromagnetic wave interference, leading to poor identification results. When the RMS height exceeds one wavelength, the identification essentially fails. The correlation length of the rough coal-rock interface has a smaller impact on identification results, with the variations in interface position and backscattered wave amplitude closely following the interface undulations. In practical applications of GPR for coal-rock interface identification, it is crucial to account for the impact of rough interfaces on identification results and to perform corresponding time delay and amplitude compensation. Increasing the root mean square height of the rough air-coal wall interface causes its echo to interfere with the coal-rock interface echo at local positions, making the identification of the coal-rock interface difficult.