Many experiences in our daily lives are temporally structured, enabling prediction of events that will occur in the future. We can anticipate upcoming subway stops during our daily commute, or plan multiple steps ahead when cooking a meal we have made many times. Although sequences of events in daily life can be multiple steps long, like the stations along a subway line, it is unknown how extended temporal structure enables predictions over multiple timescales. The three studies reported in this dissertation investigate how extended temporal structure is flexibly represented in memory and in the brain to guide multistep prediction. Chapter 1 demonstrates that memory for temporal structure is enhanced with memory consolidation, enabling more efficient judgements about predictable future events over time. Chapter 2 shows that temporal structure is represented bidirectionally and hierarchically across the hippocampus and across visual regions during multistep anticipation. Finally, Chapter 3 addresses how internal models are updated when regularities in our environment change: the hippocampus rapidly reconfigures memories of temporal structure in response to learning new information, which supports the planning of novel trajectories. Together, the studies presented in this dissertation shed light on how we represent internal models of the world that span multiple timescales to guide adaptive behavior.