This study investigates how dietary patterns and scores are associated with subsequent BMI and waist:height ratio (WHtR), and how BMI and WHtR are associated with subsequent dietary patterns or scores, from 2–3 to 10–11 and 4–5 to 14–15 years of age. In the Longitudinal Study of Australian Children, height, weight and waist circumference were measured biennially in children, yielding BMI z-score and WHtR. Parents, latterly children, reported frequency of child consumption of 12–16 food/drink items during the previous 24 h. At each wave, we empirically derived dietary patterns using factor analyses, and dietary scores based on the 2013 Australian Dietary Guidelines. We used structural-equation modelling to investigate cross-lagged associations (n 1972–2882) between diet and body composition measures in univariable and multivariable analyses. Dietary scores/patterns did not consistently predict WHtR and BMI z-score in the next wave, nor did BMI z-score and WHtR consistently predict diet in the next wave. The few associations seen were weak and often in the opposite direction to that hypothesised. The largest effect, associated with each standard deviation increase in BMI in wave 5 of the K cohort (age 12–13 years), was a 0·06 standard deviation estimated mean increase in dietary score (higher quality diet) in the subsequent wave (95 % CI 0·02, 0·11, P=0·003). Associations between dietary patterns/scores and body composition were not strongly evident in either direction. Better quantitative childhood dietary tools feasible for large-scale administration are needed to quantify how dietary patterns, energy intake and anthropometry co-develop.