A structure-based approach to the definition of sequence patterns characteristic of protein domains is presented by example. The approach requires a multiple sequence alignment of a family (or set of related families) as well as at least one three-dimensional structure. The pattern derived does not merely summarize the information in the known sequences but attempts to generalize the pattern specifications based on structural insight. In this example, the pattern-driven database search identified correctly most of the known type I copper-binding domains and detected the presence of a homologous domain in a previously unknown case (CopA protein). The significance of these results is discussed.