This paper defines and nonparametrically estimates continuous and discontinuous leverage effects. I study the asymptotic properties of the estimators and show that our estimators are unbiased. I further investigate the “leverage puzzle” and argue that it comes from the large bias induced by involving variance of estimated spot volatility changes in the leverage parameter estimator. Theoretically, the proposed estimators give the statistical limits of odd functions of an Ito^ semi-martingale’s increments, as a complement to that of even function in the case of estimating integrated volatility. Empirical study shows that there are more trading days exhibiting significant discontinuous leverage effects, indicating that we should pay more attention to return and volatility cojump.