A number of recent studies have argued that the notion of faithfulness to an original should be abandoned in models of discourse presentation, and particularly in accounts of direct speech presentation. This has coincided with a shift of attention in the study of discourse presentation from written to spoken data. This article discusses the arguments that have been made against the notion of faithfulness, and proposes a context-sensitive account of this notion, and of its relation to the various clines of discourse presentation and their categories. Our account is prompted partly by the results of a corpus-based approach to the study of discourse presentation, and partly by a qualitative analysis of a set of newspaper articles on a particular news story from outside this corpus, which we undertook to provide a check on the conclusions we had reached from our corpus study. We believe that if a general account of discourse presentation is to be reached, similarities and differences across a wide range of texts and text types need to be examined. Our corpus work, which involves careful and systematic comparison of a balanced set of written fictional, news and (auto)biographical narratives, is offered as a contribution to the general account referred to above. We also believe that if such a general account or theory is to be reached, scholars will need a clearer and more consistent application of the various descriptive terms which have been used in this area of study during the 20th century, in particular (a) 'discourse', (b) 'speech', 'thought' and 'writing' and (c) 'report', 'presentation' and 'representation'. [ABSTRACT FROM AUTHOR]