Abstract Ever since the publication of Halliday and Hasan’s (1976) seminal work on cohesion, many scholars have sought to explain different aspects of this textual relation in discourse. The purpose of this paper is twofold: first, to add to the study of the interaction between lexical cohesion and coherence ( Hellman, 1995 , Hoey, 1991b , Sanders and Pander Maat, 2006 ); and second, to contribute to the exploration of lexical cohesion as a measure in generic and register analysis ( Louwerse et al., 2004 , Taboada, 2004 , Tanskanen, 2006 , Thompson, 1994 ). I present an integrated model of lexical cohesion which challenges existing proposals affording particular attention to what I call ‘associative cohesion’. Using both quantitative and qualitative methods, the adequacy of this model is tested against a 15,683 word-corpus of broadcast discussions extracted from the International Corpus of English. The analysis of 11,199 lexical ties reports repetition (59%) as the most frequent lexical cohesion device, followed by associative cohesion (24%) and inclusive relations (8.2%), which are mostly produced in remote-mediated ties (81.8%) over speakers’ turns (90.7%). These are shown to be sensitive to genre-specific factors and to collaborate in topic management processes, thereby demonstrating the descriptive potential and applicability of the framework.