Optical non-linear multimodal tomography is a powerful diagnostic imaging tool to analyse human skin based on its autofluorescence and second-harmonic generation signals. Recently, the field of clinical non-linear imaging has been extended by adding coherent anti-Stokes Raman scattering (CARS)-a further optical sectioning method for the detection of non-fluorescent molecules. However, the heterogeneity of refractive indices of different substances in complex tissues like human skin can have a strong influence on CARS image formation and requires careful clinical interpretation of the detected signals. Interestingly, very regular patterns are present in the CARS images, which have no correspondence to the morphology revealed by autofluorescence at the same depth. The purpose of this paper is to clarify this phenomenon and to sensitize users for possible artefacts. A further part of this paper is the detailed comparison of CARS and autofluorescence images of healthy human skin in vivo covering the complete epidermis and part of the upper dermis by employing the flexible medical non-linear tomograph MPTflex CARS.