This paper reports on a text-based empirical project aimed at testing and refining Leech and Short's (1981) model of speech and thought presentation. A balanced British English corpus consisting of twentieth-century prose fiction and contemporary press stories was tagged using Leech and Short's categories of speech and thought presentation as a starting point. The tagging of the corpus led to the introduction of two new categories (the narrator's report of voice and the narration of internal states), and a number of sub-types of existing categories. We define and exemplify the new categories and sub-categories, indicate their frequencies in our data, and explore their effects in different text-types. We also discuss the coding difficulties posed by anibiguities and overlaps between categories, and consider the implications of such problems and other factors for the claim that the boundaries between speech and thought presentation categories are clinal in nature. Although our research reveals a wealth of evidence to support the idea that the speech and thought presentation scale is a cline rather than a series of discrete categories, it also suggests that some category boundaries (especially those at the direct/free indirect boundary) are less clinal than others.