<p>Peak-cluster depression (PCD) is a typical karst landform in SW China and a crucial area of karst research. Notably, tilting linearly trending events (written as tilt signals in follow) often appear in profile when ground penetrating radar (GPR) with an unshielded antenna is applied to PCD. Accordingly, we consider the probable sources of the tilt signals contain both the backscatters of the surrounding hills and the reflections of the subsurface inclined surfaces. The present study is dedicated to demonstrate our consideration through numerical modeling (considering karst topography and soil conductivity). The findings of numerical analysis showed that under moderate soil conductivity conditions, the tilt signals could contain the echoes from surrounding hills and subsurface inclined interfaces. Without finding out good methods to distinguish, we wonder whether the GPR data after filtering out the tilts signals is useful to interpret the depression subsurface. Then, we chose a typical peak cluster depression in Guizhou province in SW China as a case. Apart from the traditional process, F-K filtering process was applied to filter out the tilt signals specifically. Later attribute analysis was applied as an additional process to aid in interpreting depression soil depth. The real soil depths obtained from the auger showed results highly consistent with the interpretation of the retained horizontal signals. In addition, it is evaluated that the application potential of GPR for the karst PCD study is moderate due to the fact that GPR-acquired information there can partly be utilized to interpret subsurface, while the rest is mixed or contaminated with other irrelevant information. This study is hoped to provide help for future related work, especially how to deal with the tilt signals in GPR images.</p>