Filled pauses, once thought to be an extraneous aspect of language, play an important role in communication by serving as a signal of speaker difficulty. If children can make such attributions, then their interpretation of filled pauses should be speaker-specific. Using an eye-tracking paradigm, listeners were introduced to two characters with gender-typed colour preferences. These characters instructed children to look at pink or blue objects in a display using fluent (“Look at the X”) and disfluent (“Look at thee, uh, X”) instructions. Experiment 1indicated that 5-year-olds did not make any referential predictions. With the addition of filler trials, Experiment 2 indicated that 5-year-olds and adults anticipated reference to gender-typed objects during the baseline interval (“Look at”), and disfluent instructions led listeners to amend this prediction. These results suggest that children use disfluency as a marker to adjust their speaker-specific referential predictions.