Stereo (front-back) discrimination thresholds were measured in a two-interval forced-choice paradigm for chromatic (red-green) random-dot stereograms that had all detectable longitudinal and transverse aberrations removed by low-pass filtering. The thresholds were measured as a function of the luminance ratio of the red and the green stereo elements. Although individual differences were apparent, three subjects were able to fuse all the stimuli, including those at isoluminance. A quantitative, ideal-observer analysis was used to determine the neural efficiency with which color and luminance information was used in this stereo task. For two subjects, efficiency was constant as a function of the red-to-green ratio; for the third subject, efficiency was less near isoluminance.