Drug-induced idiosyncratic hepatotoxicity continues to be an important safety issue for the pharmaceutical industry. This toxicity is due, in part, to the limited predictive nature of current pre-clinical study systems. A hypothesis was formed that treatment of existing in vitro hepatocyte cultures with drugs clinically linked to idiosyncratic hepatotoxicity would result in the release of extracellular protein biomarkers indicative of liver toxicity. To test this hypothesis, a combination of proteomic and immunological techniques were used to first identify, and subsequently verify, components of the protein-laden conditioned culture media from immortalized human hepatocytes which overexpressed cytochrome p450 3A4. These cells were treated separately with seven individual compounds made up of a combination of thiazolidinedione and l-tyrosine PPARgamma agonists and HIV protease inhibitors, plus a vehicle control (dimethyl sulfoxide). For each drug class, clinically determined hepatotoxic and non-hepatotoxic compounds were compared. Two proteins, BMS-PTX-265 and BMS-PTX-837, were reproducibly and significantly increased in the conditioned media from cells treated with each of the toxic compounds as compared to media from cells treated with the non-toxic compounds (and vehicle). This result supported the hypothesis, and so a series of successive assays (western blots and enzyme linked immunosorbent assays) were used to measure the response of these two proteins as a function of an expanded set of 20 compounds. For all 20 drugs, elevations of BMS-PTX-265 correlated exactly with the known safety profile; whereas changes in BMS-PTX-837 correctly predicted the safety profile in 19 of 20 drugs (one false negative). In summary, the data supports both the pre-clinical in vitro method as a means to identify new biomarkers of liver toxicity, as well as the validity of the biomarkers themselves.