Chemistry has a tradition of ensuring quality through reporting properties & analysis, so every new compound (and many re-synthesised ones) must have published measurements of properties to justify their identity & purity. Because structure predicts properties, the pharmaceutical industry tests many compounds for biological activity for drug discovery. The data in these public publications is a major feedstock for the chemical information industry. Increased availability of large repositories of chemical compounds is creating new challenges and opportunities for the application of data-mining techniques to problems in chemical informatics. Applications of cheminformatics lie in performing virtual high-throughput screening to identify active compounds in a virtual library and predicting structure-activity relationships. Often chemical compounds are represented as graphs where the nodes represent atoms and the edges represent covalent bonds between them. Thus the challenge is to develop techniques capable of answering queries of interest to chemists over large chemical libraries. Such queries include similarity searches, sub-structure queries and finding motifs. Finding motifs can be investigated from two perspectives, one of finding the frequent ones, and the other of finding the significant structures. Algorithms that combine domain knowledge of chemical compounds with computer science techniques to efficiently answer these queries have been developed.