The exploitation of limestone and gypsum quarries is often complicated by the presence of waste-filled karstic zones. The capabilities of GPR to quickly delineate the economic material are studied. Surface radar data collected in a gypsum quarry are presented. These data provide limited information due to poor radar penetration and the presence of abundant diffraction patterns of ambiguous origin. Borehole GPR surveys are considered to delineate the karstic zone. Surface and borehole radar data are first synthesized numerically to seek optimal acquisition parameters. We use Carcione and Schoenberg (2000) modeling code, in which the 3D staggered grid pseudospectral operator is implemented. Modeling results show that minimally processed borehole and borehole-surfaced ata provide valuablea nd otherwise inaccessible information. The numerical simulations have also revealed limitations of the modeling code which requires further improvement.