The upper level stationary waves are defined as the deviations from longitudinal symmetry of the 250 hPa climatological monthly mean stream function. The climatological averaging period is over the years 1980 to 2000. The coupled model simulations are those of the climate of the 20th Century experiment (20C3M) prescribed by IPCC. The model results are compared to the NCEP/NCAR and ERA40 Reanalyses. The comparison shows the following. (1) The amplitude of the modeled waves in the Northern Hemisphere is generally weaker than observed; that is, modeled flow is too zonal. In the Southern Hemisphere, there are phase as well as amplitude discrepancies but no clear biases. (2) The correlation of the waves in the reanalyses and models in the Northern Hemisphere in winter averages about 0.9, but decreases sharply outside of winter. (3) For many models the correlation of the waves in the reanalyses and models outside of the wintertime Northern Hemisphere is poor, sometimes being less than 0.7. The most prominent systematic error, occurring across all models, is the underestimate of the trough/ridge in the Northern Hemisphere winter over the north Atlantic sector, 60¿W--0¿W. (4) The models having the coarsest horizontal resolution consistently underperform compared to the others. For models of horizontal resolution finer than approximately 2.5¿, the relation to horizontal resolution and fidelity of the simulation is somewhat less clear. (5) The interannual variability of the waves is consistently underestimated by almost all the models throughout the year and over the Northern Hemisphere. The differences between the models and the reanalyses are surprisingly large, given the large scale of the features diagnosed and the fairly long averaging period. Comparison to AMIP2 integrations indicates that differences in the ocean simulation in the coupled models are likely not an overwhelming influence on the agreement with reanalyses.