The pattern electroretinogram (PERG) is thought to be generated by the retinal ganglion cells. For coarse patterns, however, it has been suggested that the PERG is due to nonlinear summation of luminance responses. To test this hypothesis, we recorded the PERG in 8 patients with unilateral complete optic atrophy due to trauma or advanced glaucoma. Stimuli were phase-reversing checkerboards (7.8/s) with checks of 0.8 degrees and 15 degrees in size and with flashes. Retinal stimulation subtended 26 degrees x 34 degrees. In all 8 subjects, the PERG was greatly diminished using a check size of 0.8 degrees. With a check size of 15 degrees, the PERG was similarly diminished, while the flash responses were not reduced. Any overall luminance component of the stimulus, e.g., incomplete balance of light and dark areas, evoked strong luminance responses both in normal eyes and in eyes with optic nerve atrophy. Thus, intact ganglion cells seem to be necessary for a normal PERG regardless of the coarseness of the pattern. It is possible that different mechanisms (variable ganglion cell classes) contribute to the PERG response with different check sizes. Earlier reports, contradictory to these findings, are discussed. If the PERG reflects ganglion cell function even for large check sizes, stimulation of the ganglion cells with less optical degradation would be possible, enlarging its range of applications.