Publisher Summary In modern times, the task of organizing knowledge into systematic structures is studied by ontologists and library scientists, resulting in such well-known structures as the Dewey decimal system, the Library of Congress catalog, the AlMS Mathematics Subject Classification, and the U.S. Patent Office subject classification. Learning to assign objects to classes given examples is called classification or supervised learning. In supervised learning, the learner first receives training data in which each item is marked with a label or class from a discrete finite set. The learning algorithm is trained using this data. It is common to “hold out” part of the labeled data to tune various parameters used in the classifier. Once the classifier is trained, it is given unlabeled “test” data and has to guess the label. Supervised learning has been intensively studied for several decades in AI, machine learning, and pattern recognition, and of late in data warehousing and mining. In those domains, the data is usually more “structured” than text or hypertext. Structured data usually comes in relational tables with well-defined data types for a moderate number of columns. Furthermore, the semantic connection between these columns and the class label is often well understood. The chapter studies supervised learning specifically for text and hypertext documents. Text, as compared to structured data, has a very large number of potential features, of which many are irrelevant.