• A novel longitudinal approach to establishing belief-behaviour causal order. • Behavioural belief change was associated with change in vaccine intention/behaviour. • Cross-lagged panel model found reciprocal effect of beliefs and intention/behaviour. • Belief in vaccine effectiveness predicted subsequent vaccine intention/behaviour. • Campaigns that build confidence in vaccine effectiveness may help reduce hesitancy. Vaccine hesitancy remains a major barrier to ending the COVID-19 pandemic in the United States (U.S.) and an important target for communication interventions. Using longitudinal survey data, we examined whether baseline levels and changes in beliefs about the COVID-19 vaccines predicted change in vaccination intention/behaviour. Repeated measures were collected from a nationally representative sample of U.S. adults (n = 665) in July 2020 and April/June 2021. Linear regressions associated change in COVID-19 vaccination intention/behaviour with changes in beliefs about the COVID-19 vaccines' safety, effectiveness in protecting others from infection, and effectiveness in protecting oneself from infection. Changes in beliefs from T1 to T2 were significantly associated with change in vaccination outcomes for all belief types (safety B = 0.39, SE = 0.07; effectiveness for self B = 0.38, SE = 0.09; effectiveness for others B = 0.43, SE = 0.07). Cross-lagged models suggested a reciprocal causal relationship between pro-vaccine beliefs and vaccination intention/behaviour: Intention to get vaccinated at T1 predicted strengthened safety and effectiveness beliefs at T2. T1 effectiveness beliefs predicted T2 vaccination intention/behaviour, though T1 safety beliefs did not. Communication interventions highlighting the protective benefits of COVID-19 vaccines may be particularly successful in reducing vaccine hesitancy. [ABSTRACT FROM AUTHOR]