Rifampin is a potent inducer of CYP3A4 in vitro and precipitates numerous drug-drug interactions (DDIs) when coadministered with CYP3A4 substrates. In the current study, we have critically assessed reported rifampin in vitro CYP3A4 induction data in Fa2N-4, HepaRG, and cryopreserved or primary human hepatocytes, using either CYP3A4 mRNA or probe substrate metabolism as induction endpoints. An in vivo data base of intravenously administered victim drugs (assuming hepatic induction only) was collated (n = 18) to assess the predictive utility of these in vitro systems and to optimize rifampin in vivo E max. In addition, the effect of substrate hepatic extraction ratio on prediction accuracy was investigated using prediction boundaries proposed recently (Drug Metab Dispos 39:170-173). Incorporation of hepatic extraction ratio in the prediction model resulted in accurate prediction of 89% of intravenous induction DDIs (n = 18), regardless of the in vitro system or induction endpoint (mRNA or CYP3A4 activity). Effects of in vitro parameters from different cellular systems, and optimized in vivo E max, on the prediction of 21 oral DDIs were assessed. Use of mRNA data resulted in pronounced overprediction across all systems, with 86 to 100% of DDIs outside the acceptable prediction limits; in contrast, CYP3A4 activity predicted up to 62% of the oral DDIs within limits. Although prediction accuracy of oral DDIs was improved when using intravenous optimized rifampin E max, >35% of DDIs were incorrectly assigned, suggesting potential differential E max between intestine and liver. Implications of the findings and recommendations for prediction of rifampin DDIs are discussed. Copyright © 2011 by The American Society for Pharmacology and Experimental Therapeutics.