How does memory shape individuals' financial decisions? We find experimental evidence of a self-serving memory bias for gains and losses and show that this memory bias distorts beliefs and drives investment choices. Subjects who previously invested in a risky stock are more likely to remember positive investment outcomes relative to negative outcomes than are subjects who did not invest in the stock but only observed its outcomes. Importantly, subjects do not adjust their behavior to account for the fallibility of their memory, which leads to investment mistakes. They are likely to form overly optimistic beliefs and to re-invest in the stock even when doing so reduces their expected return. The memory bias we document is relevant for understanding how people form expectations from experiences in financial markets and, more generally, for understanding household financial decision-making.