Using high-frequency data on the cross-section of U.S. stocks, we analyze intraday return autocorrelations across a fine grid of different horizons: the term structure of intraday return autocorrelations. While average return autocorrelations are mostly negative, the degree of autocorrelation depends on the return horizon. On 15-minute horizons, return reversals are most pronounced. On sub-minute horizons, return continuations occur in relatively larger stocks, during periods of market stress, and in the first half hour of trading. Drawing on the literature, we derive and test hypotheses that link intraday return autocorrelations to major sources of market frictions and trading needs that systematically depend on past returns. We find evidence that return autocorrelations depend on the ease with which intermediaries can mean-revert their inventories, the degree of informational asymmetry, and dynamic hedge adjustments of option market makers.