Skin detection is an important preprocessing step for many applications. In some cases, reliable detection is needed under the real-world's challenging illumination conditions, that is, when the prevailing illumination does not correspond to the one used in calibration. Our particular goal in this paper is to design and study a three-sensor camera for these kinds of illumination conditions. This is done in three stages. First, a representative set of illuminants is selected from a given color temperature range. In the second stage, different illumination normalization methods are tested for the camera channel model. Simple bell-shaped sensors are tested with a chosen normalization method in the last stage. Different sensor combinations are evaluated based on their gamut ratios for skin and Munsell reflectances.