The complete physical characterization of the light impinging on any retinal location requires the specification of that light's spectral energy distribution (that is, its energies at a very large number of different wavelengths). Yet the perceived color of the surface scattering that light to the eye is representable as a single point in a color space (a) having only three dimensions and (b) in which the rectilinear wavelength continuum of spectrally pure light forms a bounding, nearly closed curve (Newton's "color circle"). Proposals to explain these radical differences between the high-dimensional space of physical light and the three-dimensional, closed space of perceived color in terms of sensory neurophysiology (for example, the existence of only three distinct classes of retinal cones and/or the opponent processing of their neuronal outputs) merely begs the question of why the sensory system is structured in that way. Ultimately, the structure of color space may have evolved as an accommodation to two fundamental and enduring facts about the world: (a) that spectral energy distributions have a linear composition and (b) that natural ambient illumination has three principal degrees of freedom-which must be compensated for in order that the perceived colors of objects remain invariant.