Operational remote monitoring of snowpack stratigraphy, melt water intrusions and their evolution with time for forecasting snowpack stability is not possible to date. Determination of the spatial variability of snowpack conditions on various scales requires a number of point measurements with various methods. These methods are either destructive or do not provide information about the internal structure of the snowpack. The application of a remotely controlled non-destructive sensor system would help to gain a higher spatio-temporal resolution about information of the snowpack. In this study we present results from upward-looking ground-penetrating radar (GPR) surveys from horizontal caves dug in the front wall of snow pits at the bottom of the snowpack. GPR data are compared with vertical profiles of snow hardness and density, obtained in the snow pit. Data were acquired in different areas with varying snow conditions with various GPR impulse systems, frequencies and polarizations. Radar experiments with high frequencies (> 1 GHz) detect internal layers in the snowpack in dry snow, but fail to provide clear reflections at the upper snow-air transition because of attenuation. In wet snow, the radar signals < 1 GHz are capable to penetrate a meter-thick snowpack and detect the snow surface, although the signal is strongly attenuated. Analysis of reflection phases and magnitudes allows interpretation of their physical origin in terms of changes in electrical permittivity. Varying antenna polarization causes a strongly different signal response, likely caused by the snow-pit wall present in our set-up. Forward calculation of density-based reflection coefficients between neighboring layers of varying hardness yields ambiguous results in terms of correspondence with observed radar reflections apart except for interferences of neighboring reflections. Moreover, we identify several pitfalls for future applications. The system set-up used here represents a basis for further developments towards a system, which is capable of improving information on the spatial and temporal snowpack characteristics. © 2009 Elsevier B.V. All rights reserved.