Does the observed relationship between mutual fund flows and recent performance represent irrational “return chasing” or rational learning about unobserved fund manager skill in the presence of decreasing returns to scale? We estimate a structural model of investor beliefs implicit in the fund flows and compare it with the rational Bayesian benchmark that estimated from past performance data. Our estimates imply that investors are more optimistic about fund manager’s average skill level than warranted by the historical data. They over-weight recent performance in a manner consistent with models based on the representativeness heuristic, yet respond slowly to changes in these beliefs, consistent with limited attention and/or informational frictions. Flows to retail funds imply more strongly biased beliefs than those to institutional funds.