Plasma polymerized thin films are of increasing importance in the field of transparent packaging. Tn particular, hydrocarbon coatings prepared by the reactive magnetron sputtering method combined with plasma-stimulated gas-phase polymerization have proved to have excellent diffusion barrier properties for gases and water vapor. However, the non-linear relationship between stretch failure and permeation properties poses a challenge when it comes to modeling the functional coatings to meet the required product specifications. An improvement in the functionality of these films has been achieved using the general regression neural network GRNN. The method employs the input parameters of the coating such as gas flow, pressure and process time. To improve the prediction performance the training input and output vectors were linearily transformed into the space defined by the principal components of the given training data. The degree of the polymerlike phase in the carbon network and its effect on the functional performance has been studied using a combination of spectroscopic methodology and functional testing. In conclusion, it can be stated that modeling of plasma processes according to the film characteristics makes possible a more controlled and systematic production of thin films.