This chapter investigates that how frequent patterns can be used to generate features. It represents definitions and preliminaries and provides the framework of pattern-based feature generation. The chapter discusses that approaches to generating patterns and provides techniques to prune large pattern sets. It also discusses strategies for constructing new features using patterns. The chapter provides applications of pattern-based feature generation for classification and clustering. As most pattern mining algorithms are designed for transactional data, vector data is usually transformed into transactional data with a discretization process for numeric features. Transactional data and vector data are usually handled in the same way as they are mutually convertible. Pattern mining is the first step for pattern-based feature generation, and it provides a set of candidate patterns for the next process. Direct contrast pattern mining directly discovers contrast patterns without generating a large number of candidates, thus it is more efficient than indirect contrast pattern mining.