This study explores the relationship between realized variance jump risk and conditional equity risk premium. Using high frequency records of the Standard & Poor’s 500 index, we construct a realized variance measure and estimate its jump component using a Heterogeneous Autoregressive model augmented by a jump component. We find strong evidence that the realized variance jump risk measure significantly relates to excess stock market returns in-sample and out-of-sample from 1998 to 2010. Further, the predictive power of the variance jump remains both statistically and economically significant after controlling for commonly-used return predictors, and is also independent from variance risk premium and price jump risk. Calibration-based evidence is also consistent with our empirical findings.