The relationship between molecular structures of drugs and their hepatotoxicity was studied by characterizing their structure in a new way and using formal concept analysis, a mathematical technique to condense knowledge into particular rules, which does not imply linearly assumptions as many conventional statistical techniques. The structural characterization was based on molecular descriptors and molecular frameworks, further decomposed into structural elements, rings, and bridges. The methodology was applied to drugs in the liver toxicity knowledge base database with the potential to cause drug-induced liver injury. Numbers of atoms and bonds along with the aromatic ratio were suitable descriptors for such drugs. The higher the number of rings and asymmetric structural elements in their terminal ring systems, the higher is the probability of hepatotoxicity. Rules were found which may help to design drugs which are unlikely to be hepatotoxic.