Current in vivo and in vitro models can't keep up with the demand for the safety assessment of large numbers of compounds emerging from high-throughput strategies. Pharmaceutical companies and start-ups are therefore building new systems that they hope will be capable of predicting the toxicity liabilities of new compounds. Cheminformatics can help week out toxic compounds at the lead selection and optimization stage; toxicogenomics may provide a toxicity diagnostic capability at all stages of drug development. For both toxicology approaches, there is not yet enough high quality data to build predictive models. Toxicology-focused cheminformatics programs attempt to consolidate data from hitherto untapped sources; toxicogenomics companies are engaged in the fussy and expensive process of manufacturing data from scratch and validating them with biological experiments.