Visual cortex (V1) neurons in cat and monkey give responses that are very variable (Tolhurst et al, 1983), limiting their ability to convey information about stimulus intensity or contrast (Tolhurst, 1989). We have tried to measure how many bits of information human observers receive about the contrast of visual stimuli, so that we can compare the performance of the whole observer with that of single neurons.Observers viewed sinusoidal gratings presented in shades of grey on a computer monitor with mean luminance of 44 cd.m^-2.Gratings had a spatial frequency of 2.67 c.deg^-1, and occupied 6 deg square of visual angle. In a given experiment, gratings of 2 to 8 different contrasts were presented many times in random sequence; overall, each contrast was presented an equal number of times. Each presentation lasted 0.5 seconds. The observer knew how many different contrasts were involved in the present experiment, and had to respond to each grating presentation by hitting an appropriate numerical key on the computer keyboard. They were told whether their responses were correct or incorrect, but they were not told the correct answer. They were allowed as much practice as they wished. The experiment conformed with local ethical committee guidelines.Contingency tables were made, showing how often the observer gave each possible response for each possible grating contrast. The Mutual Information between stimulus contrast and response was calculated (Werner & Mountcastle, 1965).When an experiment contained only 2 different contrasts, observers could identify the 2 contrasts with 100% accuracy unless the contrasts differed by less than about 6 dB. Smaller contrast differences (3 dB) could be tolerated when the 2 contrasts were close to 100%. With well-spread contrasts, observers could still perform with 100% accuracy when an experiment comprised 4 different contrasts (2 bits of information). However, with 5 or more contrasts, the contrast spacing began to decrease so that observers made mistakes of identification. The highest amounts of information conveyed in any experiment were: 2.22 bits for observer DJT and 2.37 bits for MC. Although these bit values are not large (only 4-5 discrete states), they are substantially larger than the amounts of information conveyed in 0.5 sec by single V1 neurons (0.5-1 bits; Tolhurst, 1989). More information is conveyed by V1 neurons with higher firing rate (Tolhurst, 1989), and we estimate that our psychophysical results could be explained if the gratings were encoded by populations of V1 neurons (rather than single neurons) with an overall maximal firing rate of about 140 action potentials in 0.5 s (Chirimuuta et al, 2003).