Visual feedback from the writer’s own emerging text is generally assumed to be an important component in the process of writing. For instance, in Hayes’ (1996) model it is considered to be critically involved in the subprocesses of monitoring and revision. Despite this, few empirical investigations have actually addressed this issue and in contrast to general expectations contemporary research by Oxborough and Torrance (2012) suggests that visual feedback from the text may in fact not be essential for the production of coherent texts. However, research (e.g. Johansson et al 2010) shows that writers indeed monitor their texts during writing, and that there is a relation between editing behavior and the amount of time spent looking at the text. One question is which role the visual input has for the writer, since it is does not seem to be necessary for text production, in other words: what do the writer do when she look at the text during writing. Visual feedback from an emerging text can roughly be divided into two categories of gaze behavior: fixations concurrent with keyboard typing and fixations during pauses (when no keys are struck). The current study focuses on the former and used a combination of eye-tracking and keystroke logging to collect data for 14 relatively automatized touch typists when they wrote an expository text for 30 minutes. Collapsed over all participants, this rendered 26 349 instances where a keystroke occurred concurrent with a fixation on the emerging text. Of those keystrokes 18 450 belonged to text production and 4188 to text deletions (e.g. backspaces). On average, 87 % of the text production keystrokes were performed with fixations on the word currently being typed and with a mean location of 6 characters to the left of the inscription point. For the deletion keystrokes, corresponding measures were 68 % and 6 characters to the left of the inscription point. This means that the remaining keystrokes (13 % during text production and 32 % during deletions) occurred together with more distant fixations on previous text segments. While it has been argued that fixations on the word currently being typed are related to monitoring and error correction (Torrance & Wengelin, 2010) very little is known about the role of fixations away from the word currently being typed, and even less whether the visual feedback they provide are useful at all. Qualitative explorations of those instances in the present dataset suggest that they are indeed useful and frequently appear to be associated with referential cuing and content generation.