Radar sensors are widely used in intelligent transportation systems. The performance of radar sensors in practical setups depends on many factors, such as sensor specifications, choice of locations, height and orientations of sensor installation. Currently, most of these choices are made by a "rule of thumb". In many cases, trial and error changes of settings are required and several rounds of on-site measurements and adjustments are incurred, which are both inefficient and costly. To overcome these problems, this paper reports a 3D virtual radar system that can accurately predict and evaluate the performance of actual radar sensors using a standard computer. The proposed system builds upon a 3D virtual model of real world. It allows users to freely enter specifications for radar sensors and the underlying algorithm can efficient calculate and represent graphically in the 3D model by the coverage area of the chose radar sensors at the specified location in the virtual world. Comparison with experimental data obtained using hardware testbeds shows the result obtained by the virtual radar system is accurate.