FOR the dark-adapted human observer, the absolute visual threshold has been estimated to be the effective absorption of 5–10 photons in an area covered by some 500 rods1. Although the quantum statistics of light enter as a factor which determines the frequency of seeing weak light stimuli1,2, it has been suggested that there is ‘noise’ in the visual system which ultimately limits the sensitivity of the eye3,4. Each rod in the human eye contains about 108 rhodopsin molecules, and if the ‘noise’ arises from events in the rod indistinguishable from the effects of light, there must be an extremely low probability of spontaneous change produced in any single rhodopsin molecule (or at sites within the rod disk membrane leading to an elementary voltage change in the rod). We have analysed voltage fluctuations in bipolar cells as a probe of rod activity in the dogfish retina. A component of the noise has been identified as photon noise, superimposed on dark noise arising from photon-like events. This part of the dark noise has a large temperature dependence (Q10 about 8), suggesting thermal isomerisation of rhodopsin. The rate constant, extrapolated to 37 °C, would correspond to one isomerisation in 30 s in a human rod, similar to estimates from the absolute threshold for human vision.