Achieving the ability to identify individuals who are susceptible to drug-induced liver injury (DILI) would represent a major advance in personalized medicine. Clayton et al. demonstrated that the pattern of endogenous metabolites in urine could predict susceptibility to acetaminophen-induced liver injury in rats. We designed a clinical study to test this approach in healthy adults who received 4 g of acetaminophen per day for 7 days. Urine metabolite profiles obtained before the start of treatment were not sufficient to distinguish which of the subjects would develop mild liver injury, as indicated by a rise in alanine aminotransferase (ALT) to a level more than twice the baseline value (responders). However, profiles obtained shortly after the start of treatment, but prior to ALT elevation, could distinguish responders from nonresponders. Statistical analyses revealed that predictive metabolites included those derived from the toxic metabolite N-acetyl paraquinone imine (NAPQI), but that the inclusion of endogenous metabolites was required for significant prediction. This "early-intervention pharmaco-metabonomics" approach should now be tested in clinical trials of other potentially hepatotoxic drugs.