The methods currently utilized in dentistry to detect caries lesions have their limitations and alternatives are being investigated. A promising option is tooth transillumination which is based on an increase of light scattering or light absorption in the affected tissue region. In this study transillumination applied to detect approximal caries lesions was investigated using premolar teeth containing simulated caries lesions. Cavities were drilled at the approximal surface and filled with light absorbing and light scattering fluids in different dye and particle concentrations to model successive stages of lesion progress. For light absorbing cavities the extinction as function of the decadic absorption coefficient measured at the occlusal surface could be approximated by the Lambert-Beer law (r = 0.98 +/- 0.01). For light scattering cavities the extinction as a function of the decadic reduced scatter coefficient wad fitted to a straight line (r = 0.98 +/- 0.03) for mu's (lambda = 633 nm) < 1.25 mm-1. For higher reduced decadic scatter coefficients the curves levelled off due to multiple scattering. In addition, the contribution of the dentinal cavity part to the radiance change induced by the total cavity was estimated. For light absorbing cavities illuminated with red light the average contribution was 10.5 (SD 4.2)% and for those illuminated with green light it was 1.4 (SD 0.9)%, indicating that the radiance change caused by a caries lesion is mainly determined by the enamel lesion part.