Firm-specific information can affect expected returns if it affects investor uncertainty about risk-factor loadings. We show that a stock’s expected return is decreasing in factor-loading uncertainty, controlling for the average level of its factor loading. When loadings are persistent, learning by investors can induce time-series variation in price-dividend ratios, expected returns, and idiosyncratic volatility, even when the aggregate risk-premium is constant and fundamental shocks are homoscedastic. Consistent with our predictions, we estimate that average annual returns of a firm with the median level of factor-loading uncertainty are 400 to 525 basis points lower than a comparable firm without factor-loading uncertainty.