We present here a retrospective clinical study on the detection of melanoma with a dedicated multispectral imaging system (MelaFind) that generates images of skin reflectance at 10 different wavelengths in the visible and near infrared range. The reflectance data was collected for 3609 skin lesions and analyzed with a multispectral image reconstruction algorithm that retrieves a multitude of skin parameters from multispectral data. The 6-layered skin model is used here to mimic the skin structure and the equation of radiative equation (ERT) as a light propagation model, which leads to a layered-dependent distribution of skin parameters: melanin content, blood volume fraction, and oxygen saturation (StO2). These reconstructed skin parameters are further analyzed with feature extraction and classification to evaluate the performance of the MelaFind system in terms of diagnostic accuracy. The results show that the depth-resolved skin parameters improve diagnostic accuracy with increased statistical significance (p-value).