A technique for performing wavenumber-frequency filtering on S2S forecasts is introduced and applied to forecasts from the NESM and other S2S models. The contribution of MJO time scale variability to the skill in predicting convection and winds in the tropics at weeks 1-3 is examined. Compared to global MJO indices, this approach is better able to isolate skill in predicting the MJO convective envelope and regional variations in the skill of the models at predicting the MJO. In regions of high MJO activity such as the Indian Ocean and Maritime Continent, MJO variability makes substantial contributions to the local skill at weeks 2 and 3. Differences in the geographic variability of MJO predictive skill between the boreal summer and winter are observed which are consistent with the differences in MJO activity. We also examine the improvement in skill as a function of forecast lead time that comes from a time-lagged ensemble. The local MJO skill determined from comparing observed and forecasted MJO-filtered signals is contrasted with an index constructed from the leading two empirical orthogonal functions of MJO-filtered global velocity potential. Differences in MJO skill between the S2S models are closely related to differences in predicting mid-latitude geopotential heights and conditions important to tropical cyclones (e.g., shear, mid-level humidity, and low-level vorticity).