(This book is a very interesting read and a helpful resource; I'll certainly write some more about it, later on. The only thing that's a little distracting is that you have to cut through thick jargon at times. "With the scope and slipperiness of its referential field favoring imperceptible glides, the homonymic plurality of the word fiction has notably eased the erasure of boundaries between different types of discourse." (2) Huh? So, the word 'fiction' is a little unclear and has caused some confusion between people who discuss different topics. Why not just say so? Oh well...)
First, there is a derogatory sense of 'fiction'. If you expect a truthful and careful report from someone but want to criticize what you get as widely off that standard, you might call it a 'fiction'. Sometimes, journalists or politicians are accused of producing 'fictions' that way: the projections of tax revenues from a government might be called a 'fiction' by the opposition or the press, for instance. This use of the term emphasizes the untruthful nature of fiction, the aspect of being made up, although it focuses exclusively on ways of making up things from doubtful motives. (Cohn is right, I think, to criticize this use for its underlying negativity which rubs off on many more legitimate instances of fiction. Instead of calling something fictional in this sense, we can use 'fictitious' and thus avoid conflation.)
The second sense is 'fiction' as philosophical abstraction. Philosophers have sometimes called ideas which don't reflect (in their view) an underlying reality 'fictions', when they wanted to ascribe a certain usefulness nonetheless to them. (Common sense has a corresponding notion of a 'useful illusion'.) Fictions in this sense are conceptual helping constructs with a role in explanatory discourse: they have a function in a theory which explains something. In contrast to literary fictions, they don't refer to a candidate reality, something that might be the case; they don't refer to anything at all, because they are not intended to refer to something, but to help explain something. So, 'fiction' in this sense means a totally different kind of animal than when we're talking about fiction as literary genre.
Then there is, third, 'fiction' as all literary expression at all, or even wider, as "more or less a synonym for l'imaginaire and as an antonym of le réel, referring to cultural phenomena that range from the dramas of Corneille to the palace of Versailles." (7) This broad and in itself very diverse usage characteristically regards the genres it is applied to as "expressive, ideological, or visionary genres, and [deemphasizes] their narrative structure or language." (8) It's a use that sees fiction as not necessarily, or not primarily, as narrative.
Fourth and finally, in contrast to this, 'fiction' can mean all narrative. With this goes an ideological motive: "the contemporary critique of the entire intellectual foundation of traditional historical practice—of the entire practice that is based on belief in the factuality of past events." (8) In other words: even reports intended to relate facts have to be considered as fiction, because they're narrative, and all narrative is just fiction.
Cohn herself sides with Paul Ricoeur in criticizing this fourth sense as too wide: "there are certain essential differences between narrative in history and in literature" (9), and thus the definition of 'fiction' should include only the latter. This is done by restricting 'fiction', instead of having it cover all of narrative, to only 'non-referential' narrative. (Simply put, 'non-referential' means that a narrative doesn't refer to something in the real world, but to an imaginary world; it isn't about some facts, but about imaginary events, which is exactly what distinguishes it from history.)
 Dorrit Cohn, The Distinction of Fiction, Baltimore: Johns Hopkins University Press 1999. Quoted with page numbers in the text.
 Cohn quotes Kant's Critique of Pure Reason (at 4fn9) with a remark which I think must be at A771/B799, a reference to the Vernunftbegriffe as heuristische Fiktionen; although that is in fact the only occurence of the German Fiktion in that work, and seems indeed an example for the sense Cohn has in mind here, I think the derogatory sense still outweighs it in Kant if we count in the use of Erdichtung, which appears a little more frequently.
 Which is in fact exactly the role that Kant assigns them, in that passage from his methodology chapter: "Die Vernunftbegriffe sind [...] bloße Ideen, und haben freilich keinen Gegenstand in irgend einer Erfahrung, aber bezeichnen darum doch nicht gedichtete [i.e. fictional in Cohns first sense] und zugleich dabei für möglich angenommene Gegenstände [candidate realities]. Sie sind bloß problematisch gedacht, um, in Beziehung auf sie (als heuristische Fiktionen), regulative Prinzipien des systematischen Verstandesgebrauchs im Felde der Erfahrung zu gründen." (A771/B799)
 My own use of the term throughout this blog has been close to this third sense of fiction, e.g in my survey of the forms of unreality.