This study aims to assess whether rapid weight gain in infancy is predictive of childhood obesity by establishing a new longitudinal cohort using routinely collected data. A retrospective cohort study was conducted, using routinely collected child growth data obtained from all children in Tayside, Scotland born between February 1997–February 2002. Children were weighed and measured seven times between birth and school entry. Logistic regression was used to evaluate potential risk factors for childhood obesity. Presence of obesity was defined as body mass index >95th centile (UK1990 reference data) at school entry. Rapid weight gain (RWG) was defined as conditional weight gain z score >0.67 between five time points (birth, 6–8 weeks, 8–9 months, 21–24 months and 39–42 months). Complete growth data were available for 10,438 cases (55.5% of cohort), representing 3.2% of the total Scottish population born between the above dates. Prevalence of obesity at school entry was 897/10,438 children (8.6%). Six of the 26 potential confounders considered were included in the final model. These included birth weight, prematurity, maternal smoking history, height of mother, season of delivery and pre-school parental concerns. RWG at any time from birth to 39–42 months was highly predictive of childhood obesity, the risk of which varied with exposure definition. RWG between birth and 6–8 weeks of life significantly increased risk of obesity, even in the presence of normal weight gain thereafter (adjusted OR 3.68, 95% CI 2.77–4.89, P < 0.001). Routine data confirm the importance of early childhood growth in predicting childhood obesity. This has important implications for all clinicians advising parents, and for the prioritization of NHS resources to ensure good maternal health and early infant feeding practices. It also demonstrates the need to collect such data, permitting similar studies in the future.