Landlords appear to use stale information when setting rents. Among over 43,000 California rental houses in 2018–2019, those last purchased during 2005–2007 (the peak) rent for 2–3% more than those purchased during 2008–2010 (bust). Neither house nor landlord characteristics explain this “peak-bust rental spread.” To clarify the mechanism, we test cross-sectional predictions from a simple theory of rent-setting. We find empirical support for both reference dependence and distorted beliefs. In the first, monthly payments establish (recurring) reference points, against which gains or losses are measured. In the second, past sales prices distort landlords’ current estimates of house values/rents.