A four-dimensional local ensemble transform Kalman filter (4D-LETKF) is applied to the JMA global model (GSM) at a TL159/L40 resolution to assimilate real observations including satellite radiances. It turns out that the vertical error covariance localization plays an essential role in assimilating satellite radiances. This study proposes and tests a new approach of the vertical localization, where the normalized sensitivity function of satellite sensors is used as the localization weights. With the vertical localization, AMSU-A, AMSU-B, SSM/I, TMI, and AMSR-E satellite channels are assimilated appropriately and indicate clear positive impacts. Thus, the proposed localization technique is a promising way of assimilating satellite radiances within the LETKF system.