Human colour vision depends on three classes of receptor, the short- (S), medium- (M), and long- (L) wavelength-sensitive cones. These cone classes are interleaved in a single mosaic so that, at each point in the retina, only a single class of cone samples the retinal image. As a consequence, observers with normal trichromatic colour vision are necessarily colour blind on a local spatial scale(1). The limits this places on vision depend on the relative numbers and arrangement of cones. Although the topography of human S cones is known(2,3), the human L- and M-cone submosaics have resisted analysis. Adaptive optics, a technique used to overcome blur in ground-based telescopes(4), can also overcome blur in the eye, allowing the sharpest images ever taken of the living retina(5). Were we combine adaptive optics and retinal densitometry(6) to obtain what are, to our knowledge, the first. images of the arrangement of S, M and L cones in the living human eye. The proportion of L to M cones is strikingly different in two male subjects, each of whom has normal colour vision. The mosaics of both subjects have large patches in which either M or L cones are missing. This arrangement reduces the eye＇s ability to recover colour variations of high spatial frequency in the environment but may improve the recovery of luminance variations of high spatial frequency.