Ground penetrating radar (GPR) has been used as a railway substructure investigation tool since the late 1990’s and has seen significant development since then. To use GPR as a more effective tool for substructure investigation, a GPR substructure characterization model was developed. This dissertation provides a detailed description of railway track components, track geometry, soil properties and classification and substructure design. The historical background of GPR is discussed together with GPR principles, basic GPR equations, hardware and accessories as well as GPR data collection, processing and interpretation. Other in situ investigation techniques namely the dynamic cone penetrometer (DCP), light weight deflectometer (LWD) , Pencel pressuremeter, surface wave testing, remote video monitoring (RVM), multi-depth deflectometers (MDD) and continuous track modulus measurement techniques are also discussed. A comparison between the different track investigation techniques was also done, with reference to sample rate, cost, effectiveness and value. Two sites in South Africa were selected for the investigation, one with good substructure conditions used for heavy haul coal export close to Vryheid (KN test section) and the other a general freight line with poor substructure conditions near Rustenburg (NT test section). These two sites were selected to develop a GPR substructure characterization model as they provided conditions ranging from poor to very good. This was supported by the analysis of the in situ soil sampling and testing. The calculation of the track substructure modulus from RVM deflection measurements showed three times higher values for the KN test section compared to the NT test section. The subballast and subgrade thickness, the GPR ballast fouling (GBF) index as well as the GPR moisture condition index was used for the classification ranges used in the model. The subballast and subgrade layer roughness values were calculated and used for the substructure classification. The GBF index and the GPR moisture condition roughness were used for the GPR fouling index classification. The GPR deliverables were divided into four classes (i.e. very good, good, moderate and poor). The evaluation of the characterization model showed that a traditional in situ investigation will cost approximately 3.7 times more than that of a GPR investigation. It would also take two thirds of the time to complete the GPR investigation compared to the traditional in situ investigation. The study showed that GPR can be used to develop a substructure characterization model and that it would be more cost effective and efficient than traditional in situ investigation techniques. GPR surveys provide continuous measurements of the track structure condition and can therefore provide a continuous classification unlike the discreet and fragmented nature of in situ investigations. However, in situ tests can be done at certain intervals within the GPR survey or at point where the GPR classification is not clear. The best solution for railway track characterization can therefore be obtained by using GPR and in situ classification in combination. / Dissertation (MEng)--University of Pretoria, 2012. / Civil Engineering / unrestricted