Let $(X_i,Y_i)_{i \in \mathbb{Z}}$ be a stationary sequence of ${\mathbb R}^2$-valued random variables.
To test if $X_1$ and $Y_1$ are correlated in the sense of Kendall, we propose a robust correction of the usual Kendall test, valid for a large class of dependent sequences. We also show that the condition on the dependency coefficients is optimal in a certain sense, and we illustrate our results trough different sets of simulation.