In this thesis, multivariate statistical modelling approaches are applied to longitudinal data to examine how social and non-social abilities in infancy relate to later developmental and clinical outcomes in infants at high- and low-risk for autism spectrum disorder (ASD). ASD is a heritable developmental disorder, rarely diagnosed before 2 years, and characterised by impairments in social interaction, communication and restricted and repetitive behaviours (DSM-IV-TR, American Psychiatric Association, 2000). The application of multivariate techniques to experimental, clinical and questionnaire data from 54 high-risk infants (who have an older sibling with ASD) and 50 low-risk controls, seen at 7, 13, 24 and 36 months, enables the simultaneous modelling of multiple factors in relation to ASD outcome. In Chapter 3, eye-tracking of gaze following behaviour demonstrated that infants who later develop ASD symptoms can correctly follow another person's eye-gaze, but by 13 months they show reduced looking to the gazed-at object compared to typically developing infants. Looking time is an index ofreferential understanding and was also related to children's subsequent receptive vocabulary. Chapter 4 reported that both high- and low-risk 24-montholds can apply mutual exclusivity (the principle that each object has one name) to make a word-object mapping. However, high-risk toddlers did not use social feedback to learn the word, and this difficulty was related to their lower receptive vocabularies. Further, 13 month looking time (from the gaze following task) predicted later learning from reinforcing feedback, suggesting a degree of continuity in children's social understanding across development. Finally, Chapter 5 found that social (gaze following) and non-social (disengagement) attention independently predict ASD, and while disengagement predicts looking time early in development, the measures become de-correlated over time. The findings suggest that in order to understand the variable developmental trajectories leading to ASD, multiple risk markers over time should be analysed.