P-glycoprotein (P-gp) is a multispecific transporter which has a natural detoxification role. The inhibition of substrate transport by P-gp can be presented by a number of parameters, including percentage inhibition and IC50. Inhibition constant, Ki, is believed to be a more universal parameter allowing easy comparison of data from different substrate conditions. The aim of this investigation was to use molecular descriptors of the inhibitors, docking scores, and the parameters of the probe substrate for the development of Quantitative Structure-Activity Relationships (QSARs) for the prediction of P-gp inhibition constants. QSARs were developed using a number of data mining and pre-processing feature selection methods. A chi-square based regression tree followed by a boosted trees model were the most accurate in the estimation of Ki. The selected models incorporated molecular descriptors of the inhibitors followed by the molecular descriptors of probe substrates, whereas no docking scores were selected by the models. Potent P-gp inhibitors showed higher lipophilicity and molecular weights than those molecules defined by Oprea's rule.