Biomarkers for early detection of toxicity hold the promise of improving the failure rates in drug development. In the present study, gene expression levels were measured using full-genome RAE230 version 2 Affymetrix GeneChips on rat liver tissue 48 h after administration of six different compounds, three toxins (ANIT, DMN and NMF) and three non-toxins (Caeruelein, Dinitrophenol and Rosiglitazone). We identified three gene transcripts with exceptional predictive performance towards liver toxicity and/or changes in histopathology. The three genes were: glucokinase regulatory protein (GCKR), ornithine aminotransferase (OAT) and Cytochrome P450, subfamily IIC (mephenytoin 4-hydroxylase) (Cyp2C29). RT-PCR for these three genes was performed and four additional compounds were included for validation. The quantitative RT-PCR analysis confirmed the findings based on the microarray data and using the three genes a classification rate of 55 of 57 samples was achieved for the classification of not toxic versus toxic. The single most promising biomarker (OAT) alone resulted in a surprisingly 100% correctly classified samples. OAT has not previously been linked to toxicity and cell death in the literature and the novel finding represents a putative hepatotoxicity biomarker.