The variability of the discharge of visual cortical neurons in cats and macaque monkeys limits the reliability with which such neurons can relay signals about weak visual stimuli. In general, the variance of a neuron's firing rate is directly proportional to its mean firing rate. The probability that a neuron will fire a criterion number of impulses on a stimulus trial grows monotonically with the contrast of a sinusoidal grating stimulus. Neural probability functions prepared either by computing the probability of criterion response or by integrating receiver operating characteristics to yield the probability of correct choice in a two-alternative forced-choice situation resemble psychometric functions obtained in psychophysical and behavioral experiments on humans and animals, but are shallower in slope. The slopes of neuronal probability functions are slightly higher when they are estimated over short time periods, but even so do not equal the slopes measured psychophysically in human and monkey observers. This discrepancy in slope could be explained if the whole observer responded only when about four neurons were active together.