We examine the predictability of realized measures on the cross-section of stock returns around earnings announcements. We construct realized measures (variance, skewness, kurtosis, and relative jump) using high-frequency intraday stock prices. Our results show that realized variance, skewness, and relative jumps strongly predict stock returns around earnings announcements but realized kurtosis does not. These findings are robust to various event windows, after controlling for firm characteristics, and to a range of additional tests. We further show that the predictability of realized measures is not affected by unexpected earnings. The findings also suggest that pre-announcement realized measures absorb part of the information contained in unexpected earnings.