The impact of air-sea coupling on the dynamics of the tropical Madden-Julian oscillation (MJO) is investigated with an atmospheric general circulation model (GCM) coupled to an ocean mixed layer model. In the uncoupled GCM, where climatological sea surface temperature (SST) is specified, realistic space-time spectra of near-equatorial zonal wind and precipitation are produced. However, the simulated MJO is more than twice as strong as observed, largely resulting from enormous activity during northern summer. Furthermore, during southern summer when the observed MJO is most dominant across the Indian and west Pacific Oceans, intraseasonal variance in the uncoupled model is overly concentrated to the north of Australia with little activity extending into the equatorial Indian and Pacific Oceans. Contrary to other recent modeling studies, coupling did not alleviate either of these problems nor did it have any other appreciable impact on the models MJO. Coupling has no impact because little in the way of sea surface temperature anomalies are produced by the models MJO. The lack of any significant SST anomalies stems from the lack of large-scale, spatially coherent surface heat flux anomalies driven by the models MJO. While the model does produce realistic short wave radiation anomalies, its latent heat flux anomalies are too weak, not spatially coherent, and do not constructively add with the short wave radiation anomalies. These deficiencies stem from the lack of mean westerlies across the warm pool and unrealistic phasing of surface westerlies relative to enhanced equatorial convection. It is concluded that coupling is not a panacea for problems of simulating the MJO in uncoupled GCMs and that coupling, if it does occur, results from subtleties of the structure of the MJO.