Visual fixation, the act of maintaining the eyes directed toward a location of interest, is a highly skilled behavior necessary for high-level vision in primates. In spite of its significance, visual fixation is not well understood; it is not even clear what attributes of the visual input are used to control fixation. Here we show, in four Macaca fascicularis monkeys, that the position the eyes assume during fixation depends on the luminance of the background. Dark background yields fixation positions that are shifted upward with respect to the fixation positions obtained with a dimly illuminated, featureless background. This phenomenon was observed previously in a nutshell by Snodderly; here first we rigorously establish its existence by testing proper controls. We then study the properties of this upshift of the fixation position. We show that, although the size of the upshift varies between monkeys, for all monkeys the upshift is larger than the radius of the fovea. Hence, if the background is dim, the eyes are positioned during fixation so that the target does not fall on the fovea. The size of the upshift remains almost unchanged while the eyes fixate at different orbital positions; thus the upshift is not caused by orbital mechanics. The upshift clearly is present even at the first days of training, but with additional training in fixation with dark background, the upshift increases in size. The upshift rotates with the head. The upshift increases gradually with decreasing levels of background luminosity. Luminosity, not visual contrast, is indeed the primary variable determining the size of the upshift. The contribution of a unit area of the retina to the upshift decreases as inverse square root of distance from the target; therefore, it is the perifoveal region of the retina that mostly contributes to the upshift, while the far periphery has little influence. The upshift can be induced or be canceled in the midst of a fixation by changing the background illumination; hence, the upshift is indeed an attribute of the fixation control system. Finally, the fixation-upshift studied here is different from a previously reported upshift of the endpoints of memory-guided saccades with respect to their target locations. These two types of upshift add up to each other. In discussing the function of the upshift, we note a possible morphological analogue with the retinal rod distribution. The upshift moves the line of gaze to a point intermediate between the fovea and the "dorsal rod peak." The upshift thus may improve visual acuity in scotopic conditions. The brain structure in which the upshift is generated must be involved in both ocular control and visual sensation. We consider several possibilities, of which we regard as the most likely the cerebellum and superior colliculus.