Dear Effie Marcoulaki,

Some articles that may be of help are as follows.

http://www.sciencedirect.com/science/article/pii/S1385894799001734

http://www.sciencedirect.com/science/article/pii/S0378381201007609

http://www.sciencedirect.com/science/article/pii/S0378381208000411

http://www.sciencedirect.com/science/article/pii/S1876107012002192

Sorry to dissagree. Above references are not the best available (putting it mildly...) There is no theoretical reason why a group contribution method sshould apply to viscisity. Hence it should be avoided. You can try a) corresponding states, which still is debatable - but the group at NIST have done an excellent job with REFPROP or even the older SUPERTRAPP, or b) schemes based on density and distincion between polar and non polar molecules, e.g. the old work of Thodos.You could even have a look at my book "Thermophysical Properties of Fluids. An Introduction to their prediction" Imperial College Press (1996), but do not try group contribution...

Thank you very much for these references Athanasie. I also found http://dx.doi.org/10.1021/ie071572w. So there a lot to choose from! Thanks!

Dear Marc, I agree that group contribution methods are not the most reliable option, especially for viscosities. On the other hand, I need a molecular structure- based model to include in my molecular design code. So, I can't use property databases, experimental data or coefficients for specific molecules.

  Dear Effie,

While I agree with the reasons given by Marc, you could try the UNIFAC model, which in some cases gives very good results

Thank you for your response Ana. From my understanding, UNIFAC (or other activity coefficient models) are used to predict mixture viscosities once the pure component values are available.