Modern image compression codecs produce embedded bit-streams with optimal rate-distortion properties. For natural imagery the distortionmeasure should by extended by a model of human color vision. The choice of the color space becomes here crucial since it has to be optimal in terms of compressibility and in terms of its visual distortion prediction properties. For three alternative color spaces, incorporated into a wavelet-codec (JPEG 2000), the quality-equivalent compression ratios are determined by subjective testing. For each color space the contrast sensitivity is measured and incorporated. Finally objectively measurable signal properties are aligned to the results to understand better the observed performance.