To create not a text-bound but a reader-based procedure for identifying global coherence within a discourse, five major writing tasks of 13 college students enrolled in a basic skills composition course were evaluated by three readers--doctoral students and faculty members in English and Reading Education. After choosing the five most and the five least coherent papers, the readers summarized the main ideas of these texts and identified the statements contributing to and detracting from their main ideas. Analysis of the readers' findings revealed no significant differences between highly coherent and highly incoherent papers in the percentage of statements contributing to global meaning. It did show, however, that the globally incoherent texts had a higher percentage of local incoherent statements and more frequently required readers to generate major inferences from these statements. Results indicated little reader agreement either on which propositions in the more incoherent papers were significant or irrelevant, or on the global meanings of any papers, coherent or not. The findings suggest that more coherent texts supply readers with a greater number of multiple cues for the building of meaning macrostructures and thus require fewer processing resources on the part of the reader. Because reading is a transactive process, however, the existence of multiple signs does not guarantee a shared understanding of even the most coherent texts. (MM)