The predictive utility of two in vitro methods (empirical IC(50)-based and mechanistic k(inact)/K(I)) for the assessment of time-dependent cytochrome P450 3A4 (CYP3A4) inhibition has been compared. IC(50) values were determined at multiple pre-incubation time points over 30 min for five CYP3A4 time-dependent inhibitors (verapamil, diltiazem, erythromycin, clarithromycin, and azithromycin). The ability of IC(50) data obtained following pre-incubation to predict k(inact)/K(I) parameters was investigated and its utility was assessed relative to the conventional k(inact)/K(I) model using 50 reported clinical drug-drug interactions (DDIs). Models with either hepatic or hepatic with intestinal components were explored. For low/medium potency time-dependent inhibitors, 81% of the predicted k(inact)/K(I(unbound)) from IC(50) data were within an order of magnitude of the actual values, in contrast to 50% of potent inhibitors. An underprediction trend and > 50% of false-negatives were observed when IC(50) data were used in the DDI hepatic prediction model; incorporation of the intestine improved the prediction accuracy. On the contrary, 86% of the DDI studies were predicted within twofold using k(inact)/K(I) mechanistic approach and the combined hepatic and intestinal model. Use of the empirical IC(50) approach as an alternative to the mechanistic k(inact)/K(I) model for in vivo DDI prediction is limited and is best restricted to preliminary investigations.