The pattern electroretinogram (PERG), an indicator of retinal ganglion cell (RGC) function, comprises a P50 and an N95 component. We addressed the question of whether the N95 originates, like the P50, from the RGC bodies or from the change of axon orientation at the optic nerve head (ONH). Thus, we recorded multifocal PERGs for 36 retinal locations in 21 participants. Second-order kernel responses were analyzed for the dependence of peak time topography on retinal fiber lengths to the ONH separately for the positive and negative excursions. We found that peak times were longer for macular [P1 (P50-like): 50 ms; N2 (N95-like): 76)] than for peripheral responses [P1: 43; N2: 66]. For the N2 another factor was necessary to explain the variability: The time difference (deltaT: N2 minus P1) was found to be proportional to fiber length from ganglion cell body to the ONH. We calculated retinal fiber length using an analytical function by Jansonius et al. (2009, 2012) and found that a linear model with factors eccentricity and fiber length explained 82% of the total N2 time variance (p«0.001). The conduction speeds of the retinal axons were estimated from deltaT to range from 0.5 to 3.0 m/s for parafovea and periphery, respectively. The dependence of deltaT on the distance from ganglion cell body to the ONH suggests that the N2 originates at the ONH rather than at the ganglion cell body. While the multifocal N2 peaks earlier (≈76 ms) than the non-multifocal PERG-N95 (≈95 ms), considerations of high-pass filtering and frequency dependence of the mfPERG-N2 suggest that the source separation (P50 = ganglion cell body vs. N95 = ONH) also holds for the non-multifocal PERG.