Vision begins with photoisomerization of visual pigments. Thermal energy can complement photon energy to drive photoisomerization, but it also triggers spontaneous pigment activation as noise that interferes with light detection. For half a century, the mechanism underlying this dark noise has remained controversial. We report here a quantitative relation between a pigment's photoactivation energy and its peak-absorption wavelength, lambda(max). Using this relation and assuming that pigment activations by light and heat go through the same ground-state isomerization energy barrier, we can predict the relative noise of diverse pigments with multi-vibrational-mode thermal statistics. The agreement between predictions and our measurements strongly suggests that pigment noise arises from canonical isomerization. The predicted high noise for pigments with lambda(max) in the infrared presumably explains why they apparently do not exist in nature.