With the number of new drug candidates increasing every year, there is a need for high-throughput human toxicity screenings. As the liver is the most important organ in drug metabolism and thus capable of generating relatively high levels of toxic metabolites, it is important to find a reliable strategy to screen for drug-induced hepatotoxicity. Microarray-based transcriptomics is a well-established technique in toxicogenomics research and is an ideal approach to screen for drug-induced injury at an early stage. The aim of this study was to prove the principle of classifying known hepatotoxicants and nonhepatotoxicants using their distinctive gene expression profiles in vitro in HepG2 cells. Furthermore, we undertook to subclassify the hepatotoxic compounds by investigating the subclass of cholestatic compounds. Prediction analysis for microarrays was used for classification of hepatotoxicants and nonhepatotoxicants, which resulted in an accuracy of 92% on the training set and 91% on the validation set, using 36 genes. A second model was set up with the goal of finding classifiers for cholestasis, resulting in 12 genes that appeared capable of correctly classifying 8 of the 9 cholestatic compounds, resulting in an accuracy of 93%. We were able to prove the principle that transcriptomic analyses of HepG2 cells can indeed be used to classify chemical entities for hepatotoxicity. Genes selected for classification of hepatotoxicity and cholestasis indicate that endoplasmic reticulum stress and the unfolded protein response may be important cellular effects of drug-induced liver injury. However, the number of compounds in both the training set and the validation set should be increased to improve the reliability of the prediction.