Human colour vision depends on three classes of cone photoreceptors, those sensitive to short (S), medium (M) or long (L) wavelengths, and on how signals from these cones are combined by neurons in the retina and brain. Macaque monkey colour vision is similar to human, and the receptive fields of macaque visual neurons have been used as an animal model of human colour processing. P retinal ganglion cells and parvocellular neurons are colour-selective neurons in macaque retina and lateral geniculate nucleus. Interactions between cone signals feeding into these neurons are still unclear. On the basis of experimental results with chromatic adaptation, excitatory and inhibitory inputs from L and M cones onto P cells (and parvocellular neurons) were thought to be quite specific (Fig. 1a). But these experiments with spatially diffuse adaptation did not rule out the 'mixed-surround' hypothesis: that there might be one cone-specific mechanism, the receptive field centre, and a surround mechanism connected to all cone types indiscriminately (Fig. 1e). Recent work has tended to support the mixed-surround hypothesis. We report here the development of new stimuli to measure spatial maps of the linear L-, M- and S-cone inputs to test the hypothesis definitively. Our measurements contradict the mixed-surround hypothesis and imply cone specificity in both centre and surround.