Unlike equity returns, many fixed-income return measures appear to display long memory. We show that the extent of long memory differs strongly for gross and excess holding period returns on U.S. Treasury securities. Granger and others have argued that long memory may only reflect infrequent structural breaks. We explore the impact of structural instability on tests for long memory and find only weak indications that it lies behind the long memory in our return series. The evidence of long memory remains strong for yield and term premia series even after accounting for a series of potential underlying structural changes.