The paper presents a one-factor affine model of the term structure of Libor rates with autocorrelated measurement errors. It can be viewed as a central tendency model, with the theoretical arbitrage-free rates serving as stochastic means to which the observed rates revert. Two estimation techniques are compared, one based on a no-measurement-error assumption, the other on Kalman filtering. The estimates are then used in standard yield spread regressions with a view to accounting for the departure of future short rates from what the expectations hypothesis would predict.