This study assessed the reciprocal predictive relation between multiple domains of health and academic performance (AP) in elementary schoolchildren through autoregressive models. A school-based longitudinal observational study was carried out, with a baseline measurement in 2014 and a follow-up at 12 and 24 months, in an Argentine sample of 533 children aged 9-13 years. Health was measured by self-administration of the KIDSCREEN-52 child version in the classroom, which evaluates 10 dimensions of health. AP was established with the final grades in language and mathematics obtained from the school records. Other variables collected were sex, mother's educational level, socioeconomic status, and intellectual capability. To examine the interdependencies between different health domains and AP, autoregressive and cross-lag coefficients between adjacent time points were calculated with structural equation modeling. A substantial portion of AP individual differences was explained by AP in the previous wave ([beta] above 0.790). Conversely, health measures were not as stable between different waves ([beta] around 0.400). Cross-lag path coefficients showed a significant dependence on health of previous-year AP, as can be seen in the physical well-being, psychological well-being, school environment, and social acceptance dimensions of health. AP also was influenced by previous-year health, as occurs with the self-perception, mood and emotions, and school environment health dimensions that significantly predict the next AP. These results are consistent with a positive two-way association between health and AP in childhood. This analysis is important to identify priority areas of intervention that could to improve both children's AP and well-being.