We propose a modified mutual information measure to capture general asymmetric dependence between two random variables. Based on this measure, we propose a test of asymmetric dependence and examine its finite-sample performance. We show that our test has better power than competing tests with alternative dependence measures. Using the new test, we find significant asymmetric dependence in returns of commonly used stock portfolios and the market return both in the US and other developed countries. Further, the dependence between developed country markets and the US market is stronger when both markets are in a downturn.