Interface effects are often neglected when interpreting ellipsometric data on thin organic films on solid substrates. We find that on gold substrates such effects give rise to spurious absorption seen as a nonzero imaginary part of the optical response function of adsorbed protein layers. With spectroscopic ellipsometry it is revealed that these effects are more pronounced in the visible part of the spectrum than in the ultraviolet part. We conclude that the standard three-phase model (ambient-layer-substrate) in general is insufficient for an accurate analysis of the layer properties. Large errors in the deduced refractive indices, film thicknesses, and surface concentrations may arise. A four-phase model for interpretation of spectroscopic ellipsometric data on protein layers on gold is proposed as an improvement. In this model the interaction between the adsorbed layer and the gold substrate is included and modeled in effective medium approximations. The physical mechanism behind the interaction is suggested to be depletion of electrons from a small zone in the gold substrate, close to its surface. By using this new model, a considerable improvement is obtained and the spectra obtained are free from the unphysical absorption in the visible part of the photon energy spectrum. The time development of the interaction between the protein layer and the gold substrate leads to a suggestion of a two-state adsorption model for protein adsorption. The studies in this report are limited to layers of ferritin and albumin, but we think that the results are applicable to other protein layers as well.