Riemannian geometry has been shown to significantly improve BCI classification performance. However, BCIs are still not reliable enough. To further improve Riemannian BCIs, it is thus worth exploring complementary features to the conventional Riemannian feature, i.e., spatial covariance matrix. In this work, we propose to combine the phase and amplitude covariance (PAC) of cross-frequency bands (FBs) as such an additional feature, inspired by phase-amplitude coupling.