Observer perceptions of skin color with a 15-step skin tone panel were evaluated during an as yet unpublished case-control study of malignant melanoma. Skin color is a risk factor for melanoma, and the skin tone panel was introduced in an effort to reduce its misclassification. Reflectances of the 15 artificial skin tones were measured at four wavelengths with a reflectance spectrophotometer. Six observers each evaluated the skin color of eight study subjects twice under three lighting conditions, and the results were transformed to reflectance values. Components of variance analysis demonstrated that between-subject variability contributed 63% or more of the variance at wave-lengths of 400-600 nm, while observers, light source, observation time, and error contributed 30% or less. At 700 nm, only 25.5% of the variance was due to subjects, indicating lower levels of reliability. Similarly, the correlation of visual and spectrophotometric assessment of skin reflectance was higher at 400-600 nm (r = 0.63-0.71) than at 700 nm (r = 0.41). Thus, the value of the skin tone panel-based assessments depends upon knowledge of which wavelengths most closely relate to the physiologic risk factor. For instance, reflectance at 650-700 nm is a better measure of skin melanin content than reflectance at lower wavelengths. Since the role of melanin as a risk factor remains in doubt, the utility of this technique has yet to be demonstrated. However, data from the case-control study and from this validity and reliability study will allow us to develop an analytic approach that minimizes misclassification of skin color as a confounder.