IN our attempts to unravel the workings of visual systems, we seek neural responses and interactions which can account for visual behaviour. A popular measure of visual performance is the threshold intensity at which a stimulus is detected with a given probability. To account for these detection tasks, one must not only measure neural signals at threshold1,2, but also find the limiting sources of variance (that is, the noise) both within3 and outside4,5 the nervous system6. The manner in which the random nature of photon absorptions can limit the performance of visual systems is well understood4,5,7,8. When bleaching effects are negligible, photon absorptions follow the Poisson distribution. Consequently photon counts derived from the same mean signal have a variance equal to their mean. These fluctuations in counts are an inherent property of photon signals and are referred to as photon shot noise. Processes within the visual system generate intrinsic noise7, but to assess its effect on thresholds one must first account for photon shot noise. This is generally difficult because photon catch cannot be measured directly, and must be estimated from optical parameters which are subject to significant error7. We have estimated the levels of photon shot noise and intrinsic transducer noise9 in locust photoreceptors at low light intensities. We have chosen to search for sources of intrinsic noise in the visual system of the locust because it has recently been shown that, at low intensities, each effective photon produces a single, large quantum bump10 (Fig. 1). Thus, we measure precisely the number of photons contributing to photoreceptor signals and derive, from the total response variance, that additional variance due to intrinsic noise. We report here that the transducer noise mimics the properties of photon shot noise, is of similar magnitude, and, in most conditions, must have an equally significant role in determining thresholds.