Learning which experiences reliably co-occur in time is fundamental to episodic memory and improves markedly across childhood and adolescence. Although children and adults both engage the hippocampus while learning predictable sequences, the nature of the neural representations supporting statistical learning across development remains unknown. Here, we directly quantified item-level neural representations before and after children, early adolescents, and adults learned predictable temporal relationships between items, providing a direct measure of learning-related representational change. We identified three developmental shifts in hippocampal representation. First, although posterior hippocampus integrated temporally adjacent sequence elements similarly across age groups, integration of non-adjacent sequence elements increased with age in anterior hippocampus, indicating developmental expansion in the temporal scale of neural integration. Second, hippocampal representations changed in their directional organization, with children showing hippocampal representations reflecting only forward associations between adjacent events, whereas adolescents and adults exhibited bidirectional integration of sequence relationships. Third, functional connectivity between anterior hippocampus and frontoparietal cortex tracked statistical transition probabilities during learning and predicted memory performance. Together, these findings show that improvements in statistical learning during development reflect reorganization of hippocampal representations and hippocampal-cortical interactions, revealing how the developing brain constructs increasingly flexible representations of predictive temporal structure.