We use a 20-year record of HIRS radiance measurements to evaluate the fidelity of interannual co-variability of tropical humidity and temperature in the reanalyses and GFDL AM2 simulations. Large inconsistencies between the NCEP and ECMWF reanalyses are found as are disagreements between the reanalyses and AM2 simulations. The largest discrepancies occur in the middle and upper troposphere where the NCEP and ECMWF tropical-mean relative humidity anomalies are found to be negatively correlated. When compared to HIRS dataset, NCEP is found to have unrealistically large interannual variablity in both the upper (6.7 ¿m) and middle (7.3 ¿m) tropospheric humidity channels. The radiance anomalies simulated from AM2 model output are shown to agree well with those observed by HIRS. These results support the validity of the strong coupling between temperature and humidity variations simulated in the GFDL AM2 and highlight the need to improve the representation of interannual variations of humidity in the reanalyses.