Mega-Voltage systems are used in radiation oncology both for external radiation delivery and patient positioning prior to treatment. A pair of portal images compared with digitally reconstructed radiographs is currently the gold standard for positioning but new developments have made possible the use of Mega-Voltage Cone Beam CT for better 3D setup. The non-ideal imaging geometry of the treatment unit has a direct impact on both methods. It led to the use of a reticule attachment as reference for the scale and the isocenter position on the portal images. The reticule has limited precision and occasionally super-imposes anatomical information. As for Cone Beam, the image quality crucially depends on the knowledge of the scan geometry during the acquisition. The reproducibility of the detector position at each angle will affect the image reconstruction and determine how frequently geometrical calibration must be performed. The objectives of this study are to measure the flex of the detector and evaluate its reproducibility. A RID 1640 Perkin Elmer a-Si Flat Panel is installed on a Siemens Primus linear accelerator with a positioner similar the the one used in the Oncor product. Three original methods are used to investigate the behavior in space and time of the imaging system. A reticule and a Plumb Bob tip are placed along the line formed by the isocenter and the source. Their positions projected on the flat panel for different gantry positions are used to calculate the mechanical flex. Projection matrices obtained in a geometrical Cone Beam calibration are also used to quantify the flat panel sagging. Six full sets of data were acquired over a period of 5 months and recorded overall mechanical flexes of 1 and 3 mm for the transversal and longitudinal directions respectively. The absolute magnitude of the flat panel displacement varies slightly with the method used but the discrepancy stays within the laser precision used for alignment. The small standard deviations of the flat panel displacement (< 1 mm) suggest great stability over time and permits the clinical implementation of patient positioning without the reticule. More experiments on the positioner with the complete set of projection matrices need to be performed to characterize the long-term behavior of the system and to determinate how frequently the Cone Beam calibration needs to be done to conserve image quality. Future work will develop a daily QA protocol to detect possible collisions that would bring the Cone Beam imaging system out of geometrical calibration.