We develop a framework for measuring under- and overreaction in expectation formation. The basic insight is that under- and overreaction to new information is identified (up to sign) by the impulse response function of forecast errors. Our measurement procedure yields estimates of under- and overreaction to different shocks at various horizons. In an application to inflation expectations, we find that forecasters underreact to aggregate shocks but overreact to idiosyncratic shocks. We illustrate how our approach can be used to (i) quantify the importance of different biases; (ii) estimate theoretical models; and (iii) shed light on existing empirical approaches and puzzles.