We present linear color representations for efficient color computation during image synthesis. The representations follow from describing the spectral power distributions Of all light soul ces in a scene as sums over sets of fixed basis functions. Surface and volume scattering elements are given by matrices obtained by reprojecting every scattered spectral power distribution into a user-defined subspace of the wavelength domain. This approach is a generalization of previous linear representations, such as point sampling and other matrix methods. However, because both the basis and reprojection functions can be chosen for each rendering situation, the representations provide increased accuracy and flexibility for a variety of applications. Notably, the representations can provide particularly efficient descriptions of lights and objects in local illumination renderers. Consequently, are a promising candidate for use in hardware-architecture rendering systems. We review current hardware-accelerated commonly used for color-image synthesis and describe the extensions required to employ the linear color representations. We then discuss and show examples from a prototype system that provides interactive rendering rates with accurate color computations.