The study reported in this paper applies Sinclair’s (2004) descriptive model of lexical items, which consists of five categories of co-selection: two obligatory categories the core and semantic prosody, and three optional categories collocation, colligation and semantic preference. The study examines a selection of spoken discourse events collected in Hong Kong during and in the immediate aftermath of the SARS crisis in 2003. These discourse events form part of the Hong Kong Corpus of Spoken English (HKCSE). The findings show that once the overlapping patterns of co-selection of the most frequently occurring lexical words in the SARS corpus have been determined, it is possible to describe the cumulative effects of the habitual co-selection in the lexical items that contribute to textual meanings and coherence within and across the texts. It is argued that patterns of co-selection provide a fuller picture of textual and intertextual coherence than concentrating solely on lexical cohesion.