Hidden knowledge can be represented by several kinds of knowledge artifacts, usually called patterns. Market-basket association rules, clusters, clickstreams, and image features are only some examples of significant types of patterns. Despite their heterogeneity, each pattern represents a semantic characteristic of a certain, possible huge, raw dataset in a concise way. The ability to manipulate different types of patterns under a unique environment is a fundamental issue for any data-intensive application. Unfortunately, the specific characteristics of patterns make traditional DBMSs unsuitable for pattern representation and management. Indeed, patterns can be generated from different application contexts resulting in very heterogeneous structures, possibly sharing hierarchical relationships. The aim of the project is the design of a unified framework dealing with heterogeneous patterns in a homogeneous way. The framework provides a pattern model and specific manipulation and query languages, supporting the combination of heterogeneous patterns together and with raw data. Theoretical issues concerning the expressive power and complexity of the proposed framework have also been investigated.