LAY ABSTRACT: We investigated whether a commonly used research assessment - the Autism Observation Scale for Infants (AOSI) - accurately measures autism behaviours among infants showing early signs of autism identified within the community. The AOSI is often included in studies tracking the development of infants at increased likelihood of autism, such as the infant siblings of diagnosed children. However, the suitability of this measure has not previously been tested with community-referred infants. We administered the AOSI with infants when aged 9 to 14 months and again 6 months later. Our researchers - independent of the AOSI development team and newly trained on this measure - were able to administer the brief interactive assessment and score it accurately. The infants' AOSI scores were linked to their scores on other established and validated clinical assessments, particularly at the second visit when average age was 18 months. Stronger correspondence of AOSI and other scores at this second visit suggests early autism behaviours are better established and more consistent by 18 months of age, even though these infants showed clear enough signs of possible autism to prompt referral to our study around 12 months of age. However, the moderate association of AOSI scores over time suggests that, like infant siblings - who mostly do not develop autism - community-identified infants showing early signs may also have variable developmental pathways in early life.