Results from the S2S project (by F. Vitart) suggest that global models have skill in predicting the MJO phase with up to two weeks of lead time (four weeks for the ECMWF model). Meanwhile, our results show that the MJO-TC modulation of storm genesis is reasonably captured, with some models (e.g., ECMWF, BoM, NCEP, MetFr) performing better than others. However, we also find that the models' skill in predicting basin-wide genesis and accumulated convective energy (ACE) are mainly due to the models’ ability in simulating the climatological TC seasonality. Removing the seasonality significantly reduces the models’ skill. Even the best model (ECMWF) in the most reliable basins (western north Pacific and Atlantic) has very little skill in predicting TC anomalies (close to 0.1 in Brier skill score for genesis and close to 0 in rank probability skill score for ACE). This brings up the question: do other factors contribute to the intraseasonal TC prediction skill besides seasonality? Is the low skill, after removing the seasonality, due to poor MJO simulations, or to poor representation of some aspect of the MJO-TC relationship? We will quantitatively discuss the dependence of the TC prediction skill on MJO, focusing on Western North Pacific and Atlantic, where the S2S TC predictions are relatively more skillful. We will analyze the models' skill in predicting TC genesis, ACE, spatial distribution of daily occurrence, and their dependence on MJO phase, intensity and models' characteristics.