This study uses the Local Ensemble Transform Kalman Filter (LETKF) to assimilate AIRS specific humidity retrievals with pseudo-Relative Humidity (pseudo-RH) as observation variable. Three approaches are tested: updating specific humidity with observations other than specific humidity (“passive q”), univariate assimilation (“univariate q”) and multivariate assimilation of specific humidity (“multivariate q”), comparing for the first time the performance of uni-variate q and multivariate q with an EnKF, and exploring the impact of observations other than humidity on the humidity analysis. The results show that updating humidity analysis by either AIRS specific humidity retrievals or observations other than humidity improves the humidity and wind analyses. The improvement from multivariate q is by far the largest in both analysis time and forecast time, which is due to the interaction between specific humidity and the other dynamical variables during data assimilation, leading to more balanced analysis fields. In the univariate q assimilation, similar to that used in operational centers, the specific humidity does not interact with the other dynamical variables during data assimilation, so that the analysis increments are not physically coupled. As a result, the much larger improved humidity analysis has similar zonal wind analysis accuracy as that obtained in the passive q experiment but significant larger positive impact on 48-hour forecast wind accuracy. The 6-hour total column precipitable water forecast also benefits from the improved dynamical analysis from humidity runs, with the multivariate q experiment giving the largest improvement.