This paper examine differences between risk-neutral and objective probability densities of future interest rates. The identification and quantification of these differences are important when risk-neutral densities (RNDs), such as option-implied RNDs, are used as indicators of actual beliefs of investors. We employ a multi-factor essentially affine modeling framework apllied to German time-series and cross-section term structure data in order to identify both the risk-neutral and the objective term structure dynamics. We find important differences beetween risk-neutral and objective distributions due to risk premia in bond prices. Moreover, the estimated premia vary over time in a quantitatively significant way, which implies that the differences between the objective and the risk neutral distributions also vary over time. We conclude that one should be cautions in interpreting RNDs as representing the true expectations of market participants. The method used in this paper provides an alternative approach to identifying objective probabilities of future interest rates.