We investigated gap-junctional coupling of rods and cones in macaque retina. Cone voltage responses evoked by light absorption in neighboring rods were briefer and smaller than responses recorded in the rods themselves. Rod detection thresholds, calculated from noise and response amplitude histograms, closely matched the threshold for an ideal detector limited by quantal fluctuations in the stimulus. Surprisingly, cone thresholds were only approximately two times higher. Amplitude fluctuations in cones could be explained by a Poisson distribution of photoisomerizations within a pool of seven or more coupled rods. Neurobiotin coupling between rods and cones was consistent with our electrical recordings, with approximately six rods labeled per injected cone. The spatial distribution of tracer-coupled rods matched the light-evoked cone receptive field. The gap junction inhibitor carbenoxolone abolished both electrical and tracer coupling. Amplitude fluctuations in most rods were accounted for by the expected rate of light absorption in their outer segments. The fluctuations in some rods, however, were consistent with a summation pool of up to six rods. When single rods were injected with Neurobiotin, up to 10 rods were labeled. Rod-rod and rod-cone electrical coupling is expected to extend the range of scotopic vision by circumventing saturation at the rod to rod-bipolar cell synapse; however, because coupling also renders the rod synapse less effective at separating out photon signals from dark noise, coupling is expected to elevate the absolute threshold of dark-adapted observers.