The ability of the Community Atmospheric Climate Model, version 5 (CAM5) to simulate the Madden-Julian Oscillation (MJO) was examined in a perturbed parameter ensemble (PPE) of 1100 5-year prescribed sea-surface temperature simulations generated by altering 22 physical parameters across the parameterization suite. Commonly used metrics were used to determine which are the most sensitive parameters affecting the model's simulated MJO. Although parameters perturbed in the PPE also included those from the parameterizations of shallow convection, stratiform cloud macrophysics, and the turbulent boundary layer, the parameters from the parameterization of deep convection were found to be overwhelmingly important in affecting the MJO simulation to the exclusion of all others. The most important deep convection parameters include familiar ones like the lateral entrainment rate and adjustment time-scale but also the autoconversion of cloud water to rain and evaporation efficiency of rain suggesting that both convective cloud microphysics and macrophysics are important to low-frequency tropical precipitation variability. Although the ensemble size of 1100 is barely large enough to constrain parameter values, a new 10-year integration was performed using the indicated values of the most sensitive deep convection parameters. While the MJO metrics in this new simulation were superior to that in any member of the PPE or the control model, the mean-state of this new integration was considerably worse than the control model particularly for cloud and radiation fields which were pushed to unacceptable values. This indicates that the present version of CAM5 has difficulty in simulating the physics of the MJO while maintaining an acceptable mean state.