Calculation of layer thicknesses and optical indexes is the most important part of the spectroscopic ellipsometry experimental data analysis. Most often gradient methods like Levenberg-Marquardt (LM) are used for fitting experimental and calculated characteristics. Unfortunately gradient methods depend on starting values and often get stuck into local minimums. Problems in data fitting arise often when our knowledge about investigated structure is insufficient, e.g. substrates and layers differ from expected, chemical reactions change layers optical structure, model is complicated or contains too many parameters. Such problems, in many cases could be solved with random methods or genetic algorithm (GA) (Surf. Sci. 457 (2000) 157; J. Opt. Soc. Am. A 17 (2000) 129). In this paper combined genetic algorithm is proposed to improve ellipsometric data analysis. This algorithm uses LM method together with GA. Application of the GA in ellipsometry is illustrated with the analysis of the metaloxide- semiconductor structure. (C) 2003 Elsevier B.V. All rights reserved.