Fragment-based approaches have now become an important component of the drug discovery process. At the same time, pharmaceutical chemists are more often turning to the natural world and its extremely large and diverse collection of natural compounds to discover new leads that can potentially be turned into drugs. In this study we introduce and discuss a computational pipeline to automatically extract statistically overrepresented chemical fragments in therapeutic classes, and search for similar fragments in a large database of natural products. By systematically identifying enriched fragments in therapeutic groups, we are able to extract and focus on few fragments that are likely to be active or structurally important as scaffolds. We show that several therapeutic classes (including antibacterial, antineoplastic, and drugs active on the cardiovascular system, among others) have enriched fragments that are also found in many natural compounds. Further, our method is able to detect fragments shared by a drug and a natural product even when the global similarity between the two molecules is generally low. A further development of this computational pipeline is to help predict putative therapeutic activities of natural compounds, and to help identify novel leads for drug discovery.