In early lexical development, children must learn to map spoken words onto their respective referents. Since multiple objects are typically present when any word is used, a child is charged with the difficult task of inferring the speakers intended referent. Previous research has uncovered various cues children may use in this task, including contextual and social cues. We investigate a previously unexplored cue for inferring speaker intention during word learning: speech disfluencies. Disfluencies (such as uh and um) occur in predictable locations, such as before words that are infrequent and words that have not been previously mentioned. We conducted an eye-tracking study to investigate whether young children can make use of the information contained in disfluencies to infer a speakers intended referent. Our results demonstrate that young children (ages 2;4 to 2;8) are sensitive to disfluencies. More critically, they show that children appear to use disfluencies predictively as a cue to reference and to speaker intention as the disfluency is occurring. We also examined potential sources of learning about disfluencies in CHILDES (MacWhinney, 2000) and found that disfluencies, though rare, occur regularly and with increasing frequency over time in child-directed and child-produced speech. These results reveal that young children attend to speech disfluencies relatively early in lexical development and are able to use the disfluencies to infer speaker intention during online comprehension.