We present a detailed investigation of the refractive index of polymer:fullerene blends for photovoltaic applications. The donor polymers poly[2,7-(9,9-dioctylfluorene)-alt-5,5-(4,7-di-2-thienyl-2,1,3-benzothiadiazole)] (APFO3), poly[2,3-bis-(3-octyloxyphenyl)quinoxaline-5,8-diyl-alt-thiophene-2,5-diyl] (TQ1), and poly[2,7-(9,9-dioctylfluorene)-alt-5,5-(5,10-di-2-thienyl-2,3,7,8-tetraphenyl-pyrazino[2,3-g] quinoxaline)] (APFO-Green9) were blended with either [6,6]-phenyl-C-61-butyric acid methyl ester (PCBM) or [6,6]-phenyl-C-71-butyric acid methyl ester (PC71BM). We measured variable angle spectroscopic ellipsometry for three systems, namely APFO3:PCBM, TQ1:PC71BM and APFO-Green9:PC71BM, as a function of composition and analyze the data employing a number of models. We found that Bruggeman effective medium approximations (EMA) are not precise for the description of the optical properties of these blends. This is due to a number of reasons. First, we find that there are energy shifts associated to changes in conjugation length that cannot be accounted for using EMA. Second, blending results in a strong reduction of anisotropy. Finally, our data suggest that there is some degree of vertical segregation between components. Therefore, our results support the idea that the optical properties of polymer:fullerene mixtures should be treated as alloys rather than non-interacting blends. (C) 2014 Elsevier B.V. All rights reserved.