We examine whether a new Keynesian DSGE model can generate yields that match the spectral properties of the actual yields. For this purpose, we consider two competing specifications for utility in a DSGE model: habit formation and recursive utility. We find that the yields generated from the DSGE model with either specification for utility can match the spectral properties of the actual yields reasonably well. However, it seems fair to say that each specification has some individual disadvantages. In the case of habit formation, we need very persistent and large shocks to explain the time-varying term premia. The DSGE models with recursive utility are not free of problems, either. They have difficulty in generating either a sufficient yield curve slope or positive (negative) correlations between nominal interest rates (spreads) and output.