Agree with Balázs. While Granger is for stationary series, Toda-Yamamoto is a good choice for non-stationary ones. See http://www.sciencedirect.com/science/article/pii/0304407694016168
Thank you for your response. My query was on stationary series. Suppose, two time-series variables are stationary, but they are not cointegrated. By using Johansen and Juselius model, I have found no cointegrating relationship between these variables. In this case, is it meaningful to test Granger Causality test? Please let me know.
If two variables are not cointegrated, you can still do Granger Causality tests on the first-differenced series (if you take log transformation of the original series then first-differenced series represents growth rate/ returns of that variables). However, it would then be a short-run causality. Otherwise, you can use TY version of Granger Causality even if the series are not cointegrated. I am attaching two papers for your references. Please let me know if you need further clarification