Artificial grammar learning (AGL) is an empirical paradigm which investigates basic patternand structural processing in different populations. It can inform how higher cognitive functions, such as language use, take place. Our study used AGL to assess how children with Williams syndrome (WS) (n=16) extract patterns in structured sequences of synthetic speech, how they compare to typically developing (TD) children (n=60), and how prosodic cues affect learning. The TD group was divided into: a group whose non-verbal abilities (NVMA) were within the range of the WS group, and a group whose chronological age (CA) was within the range of the WS group. TD children relied mainly on rule-based generalization when making judgements about sequence acceptability, whereas children with WS relied on familiarity with specific stimulus combinations. The TD participants whose NVMA were similar to the WS group, showed less evidence of relying on grammaticality than TD participants whose CA was similar to the WS group. In absence of prosodic cues, the children with WS did not demonstrate evidence of learning. Results suggest that, in WS children, the transition to rule-based processing in language does not keep pace with TD children and may be an indication of differences in neuro-cognitive mechanisms.