Abstract As emerged in recent years as one of the most popular sensors for environment sensing and perception, radar has been widely used in vehicle advanced driver assistance systems (ADAS), such as adaptive cruise control (ACC), active braking control, or intelligent driving as a whole. However, the performance of radar-based applications is largely affected by the environment, which may cause significant signal attenuation with added noise or clutter. Field experiment has been adopted traditionally as one of the major, if not sole, approaches in developing and testing radar-based ADAS. However, it is hard or sometimes even impossible to set up such field environment that could cover the large variations, uncertainties and complexity of real driving scenarios. Besides high cost, long duration, lack of flexibility, and worse yet, lack of guaranteed safety, limited field experiment can hardly ensure system robustness and reliability. Virtual experiment with modeling on radar, 3D environment and targets has been proved to be an effective approach in developing and testing radar-based ADAS. The research presented in this paper will focus on estimation of one of the most important attributes on the target model, that is, radar cross section (RCS). From the mechanism of radar detection, a target reflects a limited amount of radar energy and RCS is the measure of the target reflectivity. Many different factors determine how much electromagnetic energy returns to radar, or the value of RCS, such as target material, absolute and relative size of the target in relation to radar wavelength, the incident angle at which the radar beam hits a particular portion of target, reflected angle at which the reflected beam leaves the part of the target hit, and the polarization of transmitted and the received radiation in respect to the orientation of the target. Prior research on RCS estimation has been mainly from military applications in far-field region, where the distance between radar and target are typically far from each other, and often the target is treated as a point. This is very different from ADAS in automotive applications which are mainly in near-field region where radar detection range is often within 200 meters. Therefore, the methods developed in prior art cannot be directly adopted. Although there were a few methods on near-field RCS estimation, their accuracy tends to be low. In this paper, the electromagnetic scattering mechanism is firstly analyzed with targets to be typical objects in traffic, such as cars, pedestrians, etc. Then a geometric model is developed, in which the object surfaces are divided into multiple scattering zones corresponding to different scattering mechanism. According to different surface curvature radius and scattering mechanism, the scattering zones are approximately equivalent to plane, cylinder, sphere and so on. Using the ARD model based on an improved physical optics and diffraction theory, RCS value of a zone is estimated. Then the RCS of the object surface is obtained by vector superposition of all zones. Simulation is conducted to verify the RCS estimation method. The results are compared with FEKO, an international authority of the electromagnetic simulation software, which indicates that the proposed RCS estimation method generates similar results with that from FEKO software. Thus, it is highly efficient and accurate for automotive radar modeling, and for ADAS applications.