Background: Restricted and repetitive behaviour (RRB) in autism spectrum disorder (ASD) can be assessed by different measures, which diverge in item quantity, dimensionality or source of information. However, change sensitivity has not been systematically investigated among commonly used measures, albeit its importance for clinical trials and longitudinal studies. Methods: Longitudinal data resulting from behavioural observation (Autism Diagnostic Observation Schedule-2, ADOS-2; Brief Observation of Social Communication Change, BOSCC) and parent report (Restricted Behaviour Scale-Revised, RBS-R) was collected for 134 toddlers and preschoolers aged 25–65 months diagnosed with ASD by the Autism Diagnostic Interview-Revised (ADI-R) and ADOS-2. Change sensitivity was estimated using the reliable-change index and developmental trajectories of RRB by linear mixed models and k-means clustering. Results: The RBS-R identified significantly more reliable change in RRB severity compared to ADOS-2 and BOSCC. For all measures, except the RBS-R self-injurious behaviour subscale, three distinct RRB trajectories were found as follows: increasing, stable and decreasing RRB severity. Overlap was low between trajectory group assignment across measures, as were cross-sectional correlations between ADI-R, ADOS-2, BOSCC and RBS-R. Trajectory group comparisons among measures mostly showed lower baseline RRB severity in the increasing trajectory groups and higher baseline RRB severity in the decreasing trajectory groups. The trajectory groups did not differ in age or nonverbal IQ across RRB measures, except for the RBS-R compulsive behaviour subscale, which had higher nonverbal IQ in the decreasing trajectory group. Conclusions: The dimensional questionnaire RBS-R compared to ADOS-2 and BOSCC is superior in capturing subtle changes in RRB during preschool age.