The purpose of this study was to investigate the dose response of amorphous silicon (a-Si) electronic portal imaging devices (EPIDs) under different acquisition settings for both open jaw defined fields and segmented intensity-modulated radiation therapy (IMRT) fields. Four different EPIDs were used. Two Siemens and one Elekta plus a standalone Perkin Elmer research EPID. Each was operated with different acquisition systems and settings. Dose response linearity was measured for open static jaw defined fields and 'simple' segmented IMRT fields for a range of equipment and system settings. Six 'simple' segmented IMRT fields were used. The segments of each IMRT field were fixed at 10 x 10 cm 2 field size with equal MU per segment, each field having a total of 20 MU. Simultaneous measurements with an ionization chamber array (ICA) and EPID were performed to separate beam and detector response characteristics. Three different pixel calibration methods were demonstrated and compared for an example 'clinical IMRT field'. The dose response with the Elekta EPID for 'simple' segmented IMRT fields versus static fields agreed to within 2.5% for monitor unit (MU) ≥ 2. The dose response for the Siemens systems was difficult to interpret due to the poor reproducibility for segmented delivery, at MU ≤ 5, which was not observed with the standalone research EPID nor ICA on the same machine. The dose response measured under different acquisition settings and different linac/EPID combinations matched closely (≤ 1%), except for the Siemens EPID. Clinical IMRT EPID dosimetry implemented with the different pixel-to-dose calibration methods indicated that calibration at 20 MU provides equivalent results to implementing a ghosting correction model. The nonlinear dose response was consistent across both clinical EPIDs and the standalone research EPID, with the exception of the poor reproducibility seen with Siemens EPID images of IMRT fields. The nonlinear dose response was relatively insensitive to acquisition settings and appears to be primarily due to gain ghosting effects. No additional ghosting correction factor is necessary when the pixel-to-dose calibration factor at small MU calibration method is used.