Using Consensus Economics survey data on the US 3-month bill rate and the 10 years Treasury bonds expectations for the 3- and 12-month horizons over the period November 1989 – May 2015, this article aims at testing whether a group of rational forecasters coexists with or emerges over time beside a group of forecasters employing the traditional limited information-based rules that are the extrapolative, the adaptive, the regressive and the forward-market premium rules. We estimate the time-varying weights associated with the two groups using the Kalman filter methodology and find that the aggregate expectations fail to exhibit a learning process towards rationality both for short term and long term interest rates. While long term interest rate expectations appear to be explained only by limited information rules at any time, in the case of the short term interest rate a group of rational agents seems to have operated in the market over the whole period with a small but almost constant weight simultaneously with limited information-based forecasters. Overall, for both short and long term interest rates, our results strongly suggest that experts’ forecasts are essentially based on a combination of the four traditional processes. This is consistent with the economically rational expectations theory which suggests that information costs and agents’ aversion to misestimating future interest rates determine the optimal amounts of information on which they base their expectations.