Objective: Whole-head on scalp magnetoencephalography (osMEG) is a novel, cutting-edge functional neuroimaging technique that positions MEG sensors closer to the cortical sources. OsMEG allows both for free head movements and improved spatial resolution compared to conventional MEG. OsMEG thus might improve clinical epilepsy evaluations. However, it remains largely unknown how osMEG characterizes epileptic activity. Here, we aimed to compare epileptic activity source estimation accuracy of osMEG, high-density EEG (hd-EEG), conventional MEG (convMEG) and subdural EEG (sbdEEG). Method: IED and seizure onset zone source estimations of osMEG, hdEEG, convMEG and sbdEEG were evaluated using equivalent current dipoles. Cancellation index of all non-invasive modalities were calculated and compared statistically. To further investigate any similarity between osMEG and sbdEEG, representational similarity analysis was used to compare IED source estimations of these two modalities. Results: We found that osMEG IED source estimations were significantly (p<0.05) better than both convMEG and hdEEG. Furthermore, osMEG mesial temporal lobe SOZ source estimations were superior to those of convMEG and hdEEG. OsMEG cancellation index did not differ significantly from convMEG. Interestingly, comparing osMEG and sbdEEG IED source estimation demonstrated that osMEG might be less sensitive to source directions than convMEG. Conclusion: We demonstrated that whole-head osMEG exhibited very accurate non-invasive IED and SOZ source estimations, better than both hd-EEG and convMEG. Significance: OsMEG has a promising potential to become a safe, highly sensitive neuroimaging modality for whole head epilepsy evaluations.