In medical field there is need for non-invasive diagnostic tools. One particular research area is skin cancer diagnostics. Here we study Kubelka-Munk and stochastic skin reflectance models. Our objective is to compare them to each other in terms of accuracy, usefulness and biophysical parameter retrieval using convolutional neural network. The results are promising. Both model are found suitable options for further research and used stochastic model is similar to Kubelka-Munk in terms of accuracy. In physical parameter retrieval both models perform moderately. Inverted models reasonably retrieve the pigment concentrations from the simulated test data set. With empirical testing data the inverted models are mutually consistent.