Luminance and chrominance information is extracted from natural scenes by different cone-opponent mechanisms at early stage of visual processing. How independently the two streams are then processed along the visual pathway is still a controversial issue. While some authors believe that luminance and color signals are processed separately and in parallel until high perceptual level beyond V1 (Livingstone and Hubel, 1987), others suggest that they could interact as early as the LGN (Párraga et al., 2002). Supporting the latter assumption, Párraga et al. (2002) introduced a model which includes a divisive normalization of the color-opponent channels by luminance. Here we try to access the relevance of this model when viewing natural scenes. In a 2AFC paradigm, two versions of the same image were presented: one containing only luminance and the other containing the same luminance information and a small proportion of the chromatic information from the original colored image. Participants were asked to decide which scene was colored. We found that color detection threshold increased with global luminance level. While this result seems to add further weight to the early interaction hypothesis, we have to consider a possible effect of individual differences in cone-opponent mechanisms. Assuming this variability, a contrast normalization mechanism occurring independently in luminance and chrominance channels could account for our results. Under this hypothesis, pure luminance thresholds can be predicted from the chrominance detection thresholds we measured in the first experiment. Therefore, in a second experiment, we measured detection thresholds for luminance in the same natural scenes. We found them to be much higher than expected from the independence hypothesis, leading further support to early interaction. Our findings suggest an early interaction of luminance and chrominance in the visual system and provide another argument for a nonlinear divisive normalization of color channels by luminance in natural scenes.