How does memory shape individuals' financial decisions? We find experimental evidence of a self-serving memory bias. Subjects over-remember their own positive investment outcomes and under-remember negative ones. In contrast, subjects who did not invest but merely observed outcomes do not have this bias. The memory bias affects individual beliefs and decisions to re-invest. After investing, subjects form overly optimistic beliefs about their investment and re-invest even when doing so leads to a lower expected return. The memory bias is relevant for understanding how people learn from experiences in financial markets and has general implications for individual overconfidence and risk-taking.