There has been an increasing interest in the application of electronic portal imaging devices (EPIDs) to dosimetric verification, particularly for intensity modulated radiotherapy. Although not water equivalent, the phantom scatter factor of an EPID, Spe, is generally assumed to be that of a full phantom, Sp, a slab phantom, Sps, or a mini phantom. This assumption may introduce errors in absolute dosimetry using EPIDs. A calibration procedure that iteratively updates Spe and the calibration curve (pixel value to dose rate) is presented. The EPID (Varian Portal Vision) is irradiated using a 20 x 20 cm2 field with different beam intensities. The initial guess of dose rates in the EPID is calculated from ionization chamber measurements in air, multiplied by Sp or Sps. The calibration curve is obtained by fitting EPID readings from pixels near the beam central axis and dose rates in EPID to a quadratic equation. The Spe is obtained from EPID measurements in 10 X 10 cm2 and 20 x 20 cm2 field and from the calibration curve, and is in turn used to adjust the dose rate measurements and hence the calibration curve. The above procedure is repeated until it converges. The final calibration curve is used to convert portal dose to dose in the slab phantom, using the calibrated Spe, or assuming Spe = Sp or Spe=Sps . The converted doses are then compared with the dose measured using an ionization chamber. We also apply this procedure to off-axis points and study its dependence on the energy spectrum. The hypothesis testing results (on the 95% significance level) indicate that systematic errors are introduced when assuming Spe = Sp or Spe=Sps and the dose calculated using Spe is more consistent with ionization chamber measurements. Differences between Spe and Sps are as large as 2% for large field sizes. The measured relative dose profile at dmax using the EPID agrees well with the measured profile at dmax of the isocentric plane using film in a polystyrene phantom with full buildup and full backup, for open and wedged fields, and for a broad range of field sizes of interest. The dependence of the EPID response on the energy spectrum is removed once the calibration is performed under the same conditions as the actual measurements.