The goal of this work was to develop and evaluate an end-to-end test for determining and verifying image-guided radiation therapy setup accuracy relative to the radiation isocenter. This was done by placing a cube phantom with a central tungsten sphere directly on the treatment table and offset from isocenter either by 5.0 mm in the longitudinal, lateral, and vertical dimensions or by a random amount. A high-resolution cone-beam CT image was acquired and aligned with the tungsten sphere in the reference CT image. The table was shifted per this alignment, and megavoltage anterior-posterior and lateral images were acquired with the electronic portal imaging device. Agreement between the radiation isocenter (based on the MV field) and the center of the sphere (i.e., the alignment point based on kV imaging) was determined for each image via Winston-Lutz analysis. This procedure was repeated 10 times to determine short-term reproducibility, and then repeated daily for 51 days in a clinical setting. The short-term reproducibility test yielded a mean 3D vector displacement of 0.9 ± 0.15 mm between the imaging-based isocenter and the radiation isocenter, with a maximum displacement of 1.1 mm. The clinical reproducibility test yielded a mean displacement of1.1 ± 0.4 mm with a maximum of 2.0 mm when the cube was offset by 5.0 mm, and a mean displacement of 0.9 ± 0.3 mm with a maximum of 1.8 mm when the cube was offset by a random amount. These differences were observed in all directions and were independent of the magnitude of the couch shift. This test was quick and easy to implement clinically and highlighted setup inaccuracies in an image-guided radiation therapy environment.