Geophysical methods in paleoseismology enable scientists to quickly determine shallow stratigraphy; to identify displaced, oblique layers within the fault zone; to visualize faults to a great depth and to locate trenching sites. Ground-penetrating radar (GPR) data are routinely used for such investigations. Resulting interpretation is based on subjective analyses of different and complex reflection patterns. The objective set of signs that characterize the faults on GPR profiles still does not exist. In this study we carried out GPR measurements to detect major discontinuities at several sites in Kurai fault zone - one of the most potentially earthquake-prone areas in the Altai Mountains. We used “Python-3” GPR system with 38, 50 and 100 MHz antenna units. The reflections and specific wave patterns detected on processed GPR profiles were interpreted using trenching data. Then we used the open-source modeling tool gprMax to calculate synthetic GPR data for a sequence of numerical fault models, based on trenching information. We find out which signs of fault in sediments we could detect on GPR data, and what is beyond its capabilities.