Abstract NEXUS is a computational system which uses a dictionary of 100 to 150 event/state concepts to construct representations of narrative text. Associated with each event/state concept are its deep case relations and their default values. Concepts in the dictionary are related by one of seven event/state concept coherence relations. Relationships between concepts include a list of constraints on the matching of case arguments between the two concepts. NEXUS has been successfully applied to eight paragraph-length samples of text, including “A Restaurant Story”, “The Margie Story”, and “Robbing a Liquor Store”. The resulting discourse representations have been used successfully to answer questions and compute summaries for these texts. The organization of this paper is as follows. After introducing the notion of event/state concept coherence, the paper proceeds by discussing the structure of the dictionary. This includes detailed descriptions of how individual concepts are represented and related, and some discussion of issues concerning causality and property inheritance. Then, after giving some brief examples of the representations produced by NEXUS, the NEXUS program is described. NEXUS was programmed in procedural logic in lisp, so this section will include Horn clause specifications. The fourth section of the paper shows a subportion of the dictionary and, in detail, describes NEXUS processing of five samples of text. Included in the appendix are examples of NEXUS' input and output.