This paper examines the expectations behavior of individual responses in the Survey of Professional Forecasters, the University of Michigan’s Survey Research Center survey of consumers, and the ECB Survey of Professional Forecasters. It finds that the most robust feature of all of these expectations measures is that respondents inefficiently revise their forecasts, significantly underreacting to new information. As a consequence, revisions smooth through arriving information, and expectations forget past information at a rapid rate and appear to anchor to the unconditional mean or other salient anchors. The paper then examines the micro-data evidence bearing on the hypotheses tested by Coibion and Gorodnichenko (2015), who suggest that aggregate surveys may conform to key predictions of the sticky-information model of Mankiw and Reis (2002) and/or the noisy-information model of Maćkowiak and Wiederholt (2009). This paper finds considerably less coherence with these models in the micro data. The paper also provides evidence that distinguishes this behavior from learning, suggesting that the inefficient incorporation of information is much more important quantitatively than least-squares learning in these expectations measures. Finally, this empirical regularity may bear important implications for macroeconomic dynamics, as illustrated in the last sections of the paper, as it provides a microbased foundation for an earlier paper’s finding that intrinsic persistence in expectations may be a key source of macroeconomic persistence (Fuhrer 2017). The paper sketches a model in which agents’ inefficient updating of expectations induces excess smoothness in expectations, imparting persistence to macro variables that is due strictly to the expectations formation process.