Digitised colour images quantised at 24 bits per pel represent a considerable storage requirement. For many applications a coarser quantisation is allowable or even necessary. This paper describes three variations on a method to quantise RGB data at approximately 4 bits per pel. The main criterion in the scheme is that of a minimum perceptual distortion in the reproduced data. The MacAdam chromaticity domain is used as a perceptually metric space in which equal Euclidean distances may be assumed to represent equally perceived colour differences. Histogram and cluster analysis methods are used to quantise the RGB data at 4 bits luminance and 0.75 bits chromaticity. The results appear acceptably close to the originals.