Objective. To assess associations between multiple potential predictors and change in child body mass index (BMI). Methods. In the 1997 Health of Young Victorians Study, children in Grades preparatory to three (aged 5–10 years) had their height and weight measured. Parents provided information on potential predictors of childhood overweight across six domains (children's diet, children's activity level, family composition, sociodemographic factors, prenatal factors and parental adiposity). Measures were repeated three years later in 2000/1. BMI was transformed to standardised (z) scores using the US 2000 Growth Chart data and children were classified as non-overweight or overweight according to international cut-points. Regression analyses, including baseline BMI z-score as a covariate, assessed the contribution of each potential predictor to change in BMI z-score, development of overweight and spontaneous resolution of overweight in 1 373 children. Results. BMI z-score change was positively associated with frequency of take-away food, food quantity, total weekly screen time, non-Australian paternal country of birth, maternal smoking during pregnancy, and maternal and paternal BMI. Inverse associations were noted for the presence of siblings and rural residence (all p <0.05). Predictors of categorical change (development and resolution of overweight) were less clearly identified, apart from an association between maternal BMI and overweight development (p=0.02). Multivariable models suggested individual determinants have a cumulative effect on BMI change. Conclusions. Strong short-term tracking of BMI makes it difficult to identify predictors of change. Nonetheless, putative determinants across all domains assessed were independently associated with adiposity change. Multi-faceted solutions are likely to be required to successfully deal with the complexities of childhood overweight.