The hypothesis that early-life growth patterns contribute to non-communicable diseases initially emerged from historical cohort studies, consistently associating low birth weight and infant weight gain with later disease risk. Cohort studies offer crucial life-course data on disease aetiology, but also suffer from important limitations, including the difficulty of adjusting for confounding factors and the challenge of interpreting data on early growth. Prospective randomised trials of infant diet appear to provide evidence in direct contradiction to cohort studies, associating faster early growth with disease risk. The present article attempts to resolve this contradiction on two grounds. First, insufficient attention has been directed to inconsistency of outcomes between cohort studies and prospective trials. Cohort studies can assess actual mortality, whereas prospective trials investigate proxies for disease risk. These proxies are often aspects of phenotype that reflect the 'normalisation' of metabolism in response to growth, and not all those displaying normalisation in adolescence and early adulthood may go on to develop disease. Second, a distinction is made between 'metabolic capacity', defined as organ development that occurs in early life, and 'metabolic load', which is imposed by subsequent growth. Disease risk is predicted to be greatest when there is extreme disparity between metabolic capacity and metabolic load. Whereas cohort studies link disease risk with poor metabolic capacity, prospective trials link it with increased metabolic load. Infancy is a developmental period in which nutrition can affect both metabolic capacity and metabolic load; this factor accounts for reported associations of both slow and fast infant growth with greater disease risk.