A model for estimation of activity for efavirenz analogues with the K 103N mutant of HIV-1 RT was proposed in this work. The new model is predictive that requires only variable connectivity indices in the calculations. Compared with the existing models based on Monte Carlo simulation, the new model is easier to apply with better predictive accuracy, giving an r 2 of 0.85, cross-validated Q 2 of 0.82 and average error of only 0.23 kcal/mol for the 47 efavirenz analogues concerned. This work also demonstrates that variable connectivity indices, a class of not widely recognized structural descriptors, are quite useful in the QSAR studies in the fields of pharmaceutics and biochemistry. These topological indices may play a more important role in these fields in future.