A study was designed to assess the impact that speech disfluencies such as ‘uh’ and ‘um’ have on sentence comprehension and reference resolution. From a review of the results and shortfalls of Arnold et al’s current research on the topic an eye tracking study was devised to determine whether in the presence of disfluency listeners predict discourse new and difficult to name objects, or whether they predict not to hear discourse given or easy to name objects. Participants listened to recorded pairs of statements and viewed visual displays showing images that relate to the statements. An eye tracker was used to record the proportion of fixations directed towards the object representing the conclusion of the statement (payoff object), the competitor object (another likely conclusion to the statement and shared phonetic onset with the statement’s actual conclusion) and the distracter objects (objects which were not logical conclusions to the statement but shared phonetic onset with the two logical conclusions). It was hypothesized that if the proposed dis-expectancy account were correct then the listener would look to the competitor objects and the distracter objects when disfluency is present. If Arnold’s expectancy account were correct then only the competitor objects would receive a significant proportion of fixations. Statistical analysis of the results confirmed our hypothesis that in the presence of disfluency the listener looks to everything except the discourse given or easy to name object with a significant main effect for fluency at 250ms before to 250ms after disfluency onset (F(1,23) = 9.016, p < .01), (F(1,23) = 5.830, p < .05), (F(1,23) = 5.339, p < .05) and an overall significant interaction between experimental condition and time slice (F(33, 759) = 1.556, p < .05). Significant main effects for condition and interactions between condition and time slice were found for the payoff objects (F(3, 69) = 12.385, p < .001), (F(33, 759) = 13.784, p < .001) and for the distracter objects (F(3, 69) = 8.275, p < .001), (F(33, 759) = 13.238, p < .001). The results give strong support for our dis-expectancy account where upon encountering a disfluency the listener ‘un-predicts’ given or easy to name object and looks to everything else, in this case the competitor and the distracter objects.