The aim of this work is to determine the pixel sensitivity variation and off-axis dose response of an amorphous silicon electronic portal imaging device (EPID), and develop a correction method to improve EPID dosimetry. The uncorrected or raw pixel response of the aS500 amorphous silicon EPID shows differences in response (sensitivity) of individual pixels as well as a large off-axis differential response with respect to an ion chamber in water. Both can be corrected by division of raw images by the flood-field (FF) image. However, this leads to two problems for dosimetry: (1) the beam profile is present in both the raw image and FF image, and hence is "washed out" of the corrected image, and (2) any mismatch of EPID position between dosimetry and FF calibration means that the beam profile and off-axis response in the raw image and FF are misaligned. This causes artifacts in FF division and dosimetric errors. A method was developed to measure the off-axis response and pixel sensitivity variation separately to allow correction of images at any EPID position while retaining beam profile information. The pixel sensitivity variation is applied to the imager plane and is independent of imager position. The off-axis response depends on the imager plane position relative to the beam central axis. The pixel sensitivities were derived from multiple images of the same symmetric field acquired with the detector displaced laterally between each image. The off-axis response was measured by acquiring off-axis raw images (FF correction removed) and dividing out the off-axis beam fluence and previously determined pixel sensitivity differences. The dosimetric errors due to lateral and vertical detector displacement with the conventional FF calibration method were measured and compared to the new method. Corrected EPID profiles were then compared to beam profiles measured with ion chamber in water for open fields. The EPID was found to have a large off-axis differential response with respect to an ion chamber in water, particularly for 6 MV. This increased to 13% at 15 cm off-axis for 6 MV, and 3.5% for 18 MV at the isocenter plane. The dosimetric errors introduced by detector displacement with conventional FF calibration were found to be approximately 1% per centimeter of lateral detector displacement and 0.1% per centimeter of vertical displacement. These were reduced to less than 1% for any position with the new correction method. Corrected EPID images agreed with ion-chamber measurements to within 2% (excluding penumbra and low-dose areas outside the field) for various field sizes. The new correction method gives consistent dosimetry for any EPID position and retains beam profile information in the image.