Visual evoked potential (VEP) is commonly used to evaluate visual acuity in both clinical and basic studies. Subdermal needle electrodes or skull pre-implanted screw electrodes are usually used to record VEP in rodents. However, the VEP amplitudes recorded by the former are small while the latter may damage the brain. In this study, we established a new invasive procedure for VEP recording, and made a series of comparisons of VEP parameters recorded from different electrode locations, different times of day (day and night) and bilateral eyes, to evaluate the influence of these factors on VEP in mice. Our data reveal that our invasive method is reliable and can record VEP with good waveforms and large amplitudes. The comparison data show that VEP is greatly influenced by active electrode locations and difference between day and night. In C57 or CD1 ONC (optic nerve crush) models and Brn3b(AP/AP) mice, which are featured by loss of retinal ganglion cells (RGCs), amplitudes of VEP N1 and P1 waves are drastically reduced. The newly established VEP procedure is very reliable and stable, and is particularly useful for detecting losses of RGC quantities, functions or connections to the brain. Our analyses of various recording conditions also provide useful references for future studies.