Currently, the pencil-beam redefinition algorithm (PBRA) utilizes a single electron source to model clinical electron beams. In the single-source model, the electrons appear to originate from a virtual source located near the scattering foils. Although this approach may be acceptable for most treatment machines, previous studies have shown dose differences as high as 8% relative to the given dose for small fields for some machines such as the Varian Clinac 1800. In such machines collimation-scattered electrons originating from the photon jaws and the applicator give rise to extra-focal electron sources. In this study, we examined the impact of modeling an additional electron source to better account for the collimator-scattered electrons. The desired dose calculation accuracy in water throughout the dose distribution is 3% or better relative to the given dose. We present here a methodology for determining the electron-source parameters for the dual-source model using a minimal set of data, that is, two central-axis depth-dose curves and two off-axis profiles. A Varian Clinac 1800 accelerator was modeled for beam energies of 20 and 9 MeV and applicator sizes of 15 x 15 and 6 x 6 cm2. The improvement in the accuracy of PBRA-calculated dose, evaluated using measured two-dimensional dose distributions in water, was characterized using the figure of merit, FA3%, which represents the fractional area containing dose differences greater than 3%. For the 15 x 15 cm2 field the evaluation was restricted to the penumbral region, and for the 6 x 6 cm2 field the central region of the beam was included as it was impacted by the penumbra. The greatest improvement in dose accuracy was for the 6 x 6 cm2 applicator. At 9 MeV, FA3% decreased from 15% to 0% at 100 cm SSD and from 34% to 4% at 110 cm SSD. At 20 MeV, FA3% decreased from 17% to 2% at 100 cm SSD and from 41% to 10% at 110 cm SSD. In the penumbra of the 15 x 15 cm2 applicator, the improvement was less, but still significant. At 9 MeV, FA3% changed from 11% to 1% at 100 cm SSD and from 10% to 12% at 110 cm SSD. At 20 MeV, FA3% decreased from 12% to 8% at 100 cm SSD and from 14% to 5% at 110 cm SSD. Results demonstrate that use of a dual-source beam model can provide significantly improved accuracy in the PBRA-calculated dose distribution that was not achievable with a single-source beam model when modeling the Varian Clinac 1800 electron beams. Time of PBRA dose calculation was approximately doubled; however, dual-source beam modeling of newer accelerators (e.g., the Varian Clinac 2100) may not be necessary because of less impact of collimator-scattered electrons on dosimetry.