The present study re-investigated the effect of character size on eye behaviour during reading, in order to test McConkie, Kerr, Reddix, and Zola's (1988) Saccadic Range Error (SRE) hypothesis. This assumes that saccades are biased to move the eyes a constant, optimal distance in the task (i.e., range error), while aiming at the centre of peripherally selected target words. Results showed in contradiction with this hypothesis, (1) that the linear relationship between the eye launch site and the mean landing sites in words is not invariant to character size, and (2) that the optimal launch-site distance to the centre of words varies depending on the spatial extent of the words, and differs from the mean length of saccades in the task. We propose an alternative, Center-of-Gravity hypothesis, which a priori accounts for the launch-site effect and its variations with character size, and suggests that research in reading may benefit from reconsidering the role of character size.