Over the last two decades, the UNIFAC group contribution method has been widely used for the prediction of vapor-liquid equilibria. For application to refrigerant mixtures, additional structure groups were introduced and verified by fitting the interaction parameters with satisfactory results. Analogously, the new structural groups should be implemented in the Modified UNIFAC method, which uses temperature-dependent interaction parameters and adjustable group surface area and volume parameters to achieve a better description of the behaviour of the mixtures. In order to fit these parameters to experimental data, an optimization problem with 386 variables was solved. This was done by applying Evolutionary Algorithms to mathematical optimization, involving the mutation-selection principle known from biology. The optimum interplay of many well-known strategies as well as the use of parallel computers resulted in levels well below the local extremes found using a conventional search method. The EVOBOX program package can be used for any minimization task with multivariable functions.