In this study, we used the parametric empirical Bayesian (PEB) framework for distributed source reconstruction in EEG/MEG implemented in the Statistical Parametric Mapping software (SPM) to compare forward models using real EEG data [1]. The current forward modeling in the framework comprises a 3-layered boundary element method (BEM) approximation of the head including a skull, scalp and brain compartment constructed based on a high resolution anatomical MRI template. Here we introduce forward models constructed using the finite difference reciprocity method (FDRM) [2] within the framework. This method allows using more realistic volume conductor models segmented from the template. First, we numerically verified the FDRM technique in a 3-layered spherical volume conductor model using standard metrics such as the Relative Difference Measure (RDM)[3]. After verification, two aspects were investigated: 1) the influence of using FDRM or BEM based forward modeling assuming the same three-layered default BEM volume conductor model used in SPM; 2) the effect of using a FDRM forward model including the three-layered volume conductor model extended with CSF and air cavities. For all the models we assumed the same source space. Based on 8 grand average ERP datasets across subjects and assuming multiple sparse priors [4], the models were compared using the Bayesian log- evidence corresponding with the reconstructions. Also the reconstructed activity was compared to the activity reported in previous studies of equivalent ERP waveforms. The results indicate that FDRM forward modeling and the use of a more realistic volume conductor model both improve the reconstructed activity compared to the 3-layered BEM approximation. For equal volume conductors we found strong evidence in 6 of the 8 ERP datasets in favor of the FDRM. For the extended FDRM volume conductor model we found strong evidence for all the datasets in favor of the FDRM. These results emphasize the influence of the forward model on the reconstructed activity and show that EEG source reconstruction can benefit from the use of realistic MRI based FDRM volume conductor models