Active transport of small molecules out of the cells mediated by efflux pumps of the ABC superfamily is an important biochemical phenomenon precluding delivery of drug molecules to the site of action, contributing to loss of efficacy and potential drug-drug interactions. Evaluating the efflux potential of new drug candidates is therefore crucial in the earliest stages of drug discovery. Due to its ubiquitous presence in the tissues and extremely broad substrate range, human P-glycoprotein (P-gp) has one of the largest impacts on drug distribution between tissues. A variety of computational approaches have been proposed to predict P-gp efflux, but their utility is mostly restricted to the classification of drugs into substrates and non-substrates, while the actual quantitative effect of efflux on other ADME processes remains elusive. The primary goal of the current study was to address this shortcoming by utilizing a censored-regression based statistical methodology to develop predictive models for the identification of P-gp substrates that would produce quantitative output in the form of P-gp efflux ratio (ER). The models were trained on a data set of about 3,500 compounds with ER values in exact or censored representation (i.e., only known to fall above or below a certain threshold) and a minimal selection of fundamental physicochemical descriptors, such as LogP, pKa, or McGowan Volume, enabling easy interpretation and offering mechanistic insight into the interplay between passive diffusion and active efflux processes. Moreover, we demonstrated how the described models can be used in practice to evaluate how P-gp efflux affects blood-brain barrier penetration and oral bioavailability.