We test an alternative ellipsometry data analysis technique to measure the temperature and submicroscopic surface roughness of CdTe/Ge during growth by molecular beam epitaxy. Such technique is based on a parametric model for the dielectric function and second-order differentiation of the data. It was proposed in order to reduce the effects of roughness and surface contamination, which are detrimental to the run-to-run accuracy of the traditional data analysis. We find that the new technique yields more accurate results than does the traditional analysis in the presence of both microscopic roughness and submonolayers of excess Te at the surface. However, our data indicate that the new method does not improve the run-to-run accuracy of the temperature measurements. We conjecture that this failure is due to the presence of macroscopic roughness on all MBE-grown (211) CdTe surfaces. We also propose a qualitative method to detect the formation of macroscopic roughness during growth. (c) 2006 Elsevier B.V. All rights reserved.