In heterogeneous glacial sediments, high resolution characterization of geology is crucialfor understanding contaminant transport at polluted sites. Geological information ismostly obtained from drilling of multiple investigation boreholes, but such point-baseddata alone do not always provide the required resolution to map heterogeneity betweenboreholes. Crosshole ground penetrating radar (GPR) can be used to obtain informationabout the geology between the boreholes, since the GPR signal is sensitive to changes indielectric material properties. This PhD project was initiated to develop a methodologyfor using GPR as part of contaminated site investigations.First, a robust algorithm is required that can translate GPR data into subsurface modelestimates relatively fast. Therefore, a probabilistic straight-ray-based inversion schemewas developed, where the forward modeling error arising from choosing a straight-rayforward solver is accounted for. Due to the linear formulation, we can decouple theinversion of traveltime and amplitude data and obtain results fast. The approach wasevaluated through a synthetic study, and a field study largely confirmed the results. Forboth the synthetic and field study, accounting for the forward modeling error was fun-damental to obtain tomograms without artifacts, particularly for inversion of amplitudedata.Then, the effect of water filled boreholes was investigated. Datasets obtained in boreholeswith and without water were compared to quantify the influence from water-filledboreholes in clay till environments. The results showed that the water-filled boreholescause a three-fold increase of amplitudes, and that the mean frequency of the waveformdata is reduced by approximately one third.Finally, crosshole GPR was used at an industrial contaminated site in a clay till setting andthe developed inversion approach was applied to obtain subsurface model estimates ofthe radar wave velocity and attenuation, independently. The GPR results were comparedto borehole logs, grain size analyses and relative permeability data from the site. TheGPR data analysis provided the necessary information to refine the conceptual geologicalmodel. A silt layer with a thickness of a few decimeters, likely important for flowcharacterization, was delineated by the estimated velocity model, while the attenuationmodel provided additional information about variation in subsurface clay content.