We present an experiment where subjects sequentially receive signals about the true state of the world and need to form beliefs about which one is true, with payoffs related to reported beliefs. We control for risk aversion using the Offerman et al. (2009) technique. Against the baseline of Bayesian updating, we test for belief adjustment under-reaction and over-reaction and model the decision making process of the agent as a double hurdle model where agents first decide whether to adjust their beliefs and then, if so, decide by how much. We find evidence for periods of belief inertia interspersed with belief adjustment. This is due to a combination of: random belief adjustment; state dependent belief adjustment, with many subjects requiring considerable evidence to change their beliefs; and Quasi-Bayesian belief adjustment, with insufficient belief adjustment when a belief change does occur.