Ground Penetrating Radar (GPR) is used to acquire data from near-surface depth for archeological, infrastructural, etc. researches and applications. Acquired data allow users to visualize and interpret the underground structures with high accuracy. The 3D visualization of the underground structures is one of the most problem for GPR research and applications. Usage of the suitable approach for 3D visualization will increase the accuracy of visualization and interpretation of underground structures. In order to contribute to this problem, an approach is proposed. Firstly, GPR data are acquired from the search area and data preprocessing steps are applied to GPR data. Secondly, the incomplete or missing data are recovered using interpolation techniques. Thirdly, the GPR data corresponding to the underground structures or anomalies are extracted and placed in a 3D cube. Finally, the extracted GPR data are visualized in 3D environment. The proposed approach was implemented on the real GPR data acquired from the test area. The results showed that created 3D models of the underground structures are very close to real model.