We propose to analyze ellipsometry data by using effective medium approximation (EMA) models. Thanks to EMA, having nanocrystalline reference dielectric functions and generalized critical point (GCP) model the physical parameters of two series of samples containing silicon nanocrystals, i.e. silicon rich oxide (SRO) superlattices and porous silicon layers (PSL), have been determined. The superlattices, consisting of ten SRO/SiO(2) layer pairs, have been prepared using plasma enhanced chemical vapor deposition. The porous silicon layers have been prepared using short monopulses of anodization current in the transition regime between porous silicon formation and electropolishing, in a mixture of hydrofluoric acid and ethanol. The optical modeling of both structures is similar. The effective dielectric function of the layer is calculated by EMA using nanocrystalline components (nc-Si and GCP) in a dielectric matrix (SRO) or voids (PSL). We discuss the two major problems occurring when modeling such structures: (1) the modeling of the vertically non-uniform layer structures (including the interface properties like nanoroughness at the layer boundaries) and (2) the parameterization of the dielectric function of nanocrystals. We used several techniques to reduce the large number of fit parameters of the GCP models. The obtained results are in good agreement with those obtained by X-ray diffraction and electron microscopy. We investigated the correlation of the broadening parameter and characteristic EMA components with the nanocrystal size and the sample preparation conditions, such as the annealing temperatures of the SRO superlattices and the anodization current density of the porous silicon samples. We found that the broadening parameter is a sensitive measure of the nanocrystallinity of the samples, even in cases, where the nanocrystals are too small to be visible for X-ray scattering. Major processes like sintering, phase separation, and intermixing have been revealed as a function of annealing of the SRO superlattices. (C) 2010 Elsevier B.V. All rights reserved.