Purpose: With usually a millimeter-level PTV margin, stereotactic radiosurgery (SRS) and stereotactic body radiation therapy (SBRT) pose a stringent requirement on the isocentricity of the Linac. This requirement is partly fulfilled by routine isocenter quality assurance (QA) test to verify the size and location of the isocenter. The current common QA methods such as spoke shot were developed before SBRT/SRS became popular and when IGRT was largely absent and hence have their limitations. In this work, we describe an isocenter QA approach based on portal imaging to provide the community with a superior alternative. Methods: The proposed approach utilizes a BrainLab ball bearing (BB) phantom in conjunction with an electronic portal imaging devices (EPID) imager. The BB phantom was first aligned with a calibrated room laser system. Portal images were then acquired using 6 MV beam with a 2 × 2 cm 2 open field and a 15 mm cone on a Varian TrueBeam STx machine. The gantry, collimator, and table were rotated separately at selected angles to acquire a series of portal images in order to determine the isocenter of each rotating system. The location and diameter of these isocenters were determined by calculating the relative displacement of either BB or open field edge between the acquired EPID images. The demonstration of the reproducibility and robustness of this EPID-based approach was carried out by repeating measurements 10 times independently for each rotating system and simulating clinical scenarios of asymmetric jaws and misalignment of BB phantom, respectively. Results: For our TrueBeam STx machine, the isocenter diameter derived from open-field EPID images was roughly 0.15 mm, 0.18 mm, 0.49 mm for the collimator, table, and gantry, respectively. For the collimator and gantry, images taken with the cone gave considerably smaller isocenter diameter. Results remained almost unchanged despite the presence of How to cite this paper: Fan101 Int. J. Medical Physics, Clinical Engineering and Radiation Oncology simulated BB misalignment and asymmetric jaws error, and between independent measurements. Isocenter location and diameter derived from images obtained at a limited number of angles (≤11) were adequately accurate to represent those derived from images of densely sampled angles. Conclusions: An EPID-based isocenter QA approach is described and demonstrated to be accurate, robust, and reproducible. This approach provides a superior alternative to conventional isocenter QA methods with no additional cost. It can be implemented with convenience for any linear accelerator with an EPID imager.