The single photon responses of retinal rod cells are remarkably reproducible, allowing the number and timing of photon absorptions to be encoded accurately. This reproducibility is surprising because the elementary response arises from a single rhodopsin molecule, and typically signals from single molecules display large intertrial variations. We have investigated the mechanisms that make the rod's elementary response reproducible. Our experiments indicate that reproducibility cannot be explained by saturation within the transduction cascade, by Ca2+ feedback, or by feedback control of rhodopsin shutoff by any known element of the cascade. We suggest instead that deactivation through a series of previously unidentified transitions allows the catalytic activity of a single rhodopsin molecule to decay with low variability. Two observations are consistent with this view. First, the time course of rhodopsin's catalytic activity could not be accounted for by the time required for the known steps in rhodopsin deactivation-phosphorylation and arrestin binding. Second, the variability of the elementary response increased when phosphorylation was made rate-limiting for rhodopsin shutoff.