The group contribution approach has been shown to be suited for prediction of transport parameters (gas permeability and diffusion coefficients) of glassy polymers based on the chemical structure of their repeat units. This paper presents further development of this method. An attempt is made to improve the accuracy of such predictions. A database including several hundreds of the structures of amorphous glassy polymers and their transport parameters was used for the calculations of the group contributions in the Pi and Di values. Better agreement between the experimental and predicted Pi and Di values was achieved when the data for different polymer classes (e.g. polynorbornenes and polyimides) were treated separately. The best fit of the predicted and experimental data for polyimides is obtained using a modified scheme where a distinction is made between the groups that appear in main and side chains and substituents at aliphatic and aromatic carbon atoms.