Typically developing (TD) infants enhance their learning of spoken language by observing speakers' mouth movements. Given the fact that word learning is seriously delayed in most children with neurodevelopmental disorders, we hypothesized that this delay partly results from differences in visual face scanning, e.g., focusing attention away from the mouth. To test this hypothesis, we used an eye tracker to measure visual attention in 95 infants and toddlers with Down syndrome (DS), fragile X syndrome (FXS), and Williams syndrome (WS), and compared their data to 25 chronological- and mental-age matched 16-month-old TD controls. We presented participants with two talking faces (one on each side of the screen) and a sound (/ga/). One face (the congruent face) mouthed the syllable that the participants could hear (i.e., /ga/), while the other face (the incongruent face) mouthed a different syllable (/ba/) from the one they could hear. As expected, we found that TD children with a relatively large vocabulary made more fixations to the mouth region of the incongruent face than elsewhere. However, toddlers with FXS or WS who had a relatively large receptive vocabulary made more fixations to the eyes (rather than the mouth) of the incongruent face. In DS, by contrast, fixations to the speaker's overall face (rather than to her eyes or mouth) predicted vocabulary size. These findings suggest that, at some point in development, different processes or strategies relating to visual attention are involved in language acquisition in DS, FXS, and WS. This knowledge may help further explain why language is delayed in children with neurodevelopmental disorders. It also raises the possibility that syndrome-specific interventions should include an early focus on efficient face-scanning behaviour.