Although most models of connected discourse are strikingly similar in the types of relationships they describe, they are strikingly different in two areas: the unit of analysis and the dimensions on which cohesion versus coherence are described. Common systems for analyzing written text use the sentence, T-unit, clause, or proposition as the unit of analysis. However, these units of analysis are often confounded with the issue of relationship types. That is, when the unit of analysis is large, salient relationships between unit elements are lost and when the unit of analysis is small, too many relationships are identified. A compromise unit of analysis might be predication, a system that recognizes both finite and nonfinite clauses and contains two types of relationships--argumental and nonargumental. Since the identification of referential ties (cohesion) is one of the first lines of discourse processing, it is logical to use a model that identifies lexical references from one predication to the next. Coherence can be determined by applying a set of macrorules to the discourse as a whole. For a set of predications, the macrostructure may be signaled by predications summarizing or subsuming the other predications in the set. There are also several types of patterns of relationships or macropatterns, each with governing predications. The advantage of this system of analysis is that it permits for a quantitative analysis of discourse data. (HOD)