For the interpretation of an ellipsometric measurement an adaptive optical model has to be assumed because of the lack of inverse equations. Finding the proper parameters of the optical model for multi-layer structure by minimizing the difference (error) between the measured and the computed data becomes a typical global optimization task. The technique of generalized simulated annealing (GSA) was introduced in order to solve inverse problems related to ellipsometry, and was tested in evaluating ellipsometric measurements for lead zirconate titanate films. It is shown that unlike local optimization methods, the convergence of GSA is independent of the initial model and it is capable of producing reliable results in ellipsometry with great computation efficiency.