The corpus-driven analysis of concordance data often results in the identification of groups of lines in which repeated patterns around the node item establish membership in a particular function meaning group (Mahlberg, 2005). This paper explains the KWICgrouper, a concept designed to support this kind of concordance analysis. Groups are defined by sets of patterns that can be matched against the lines in a concordance. The central elements of the KWICgrouper are described in object-oriented terms and an experimental implementation described.