It has previously been demonstrated that modeled root-zone soil moisture can be improved by assimilating remotely sensed near-surface soil moisture observations. To date most work in this area has focused on assimilating soil moisture observations derived from one of two established microwave sensors: the passive Advanced Microwave Scanning Radiometer - Earth Observing System (AMSR-E) and the active Advanced Scatteremoter (ASCAT) sensor. Due to the unique characteristics of the passive and active microwave signals, as well as the different approaches used by the retrieval algorithms typically applied to each, comparison of the soil moisture observations derived from AMSR-E and ASCAT indicates some variation in their relative skill according to local land surface conditions. Consequently, it should be possible to obtain universally more accurate near-surface soil moisture information by combining the information from the AMSR-E and ASCAT soil moisture data sets. This study pursues this approach by assimilating both AMSR-E and ASCAT soil moisture into NASA's Catchment Land Surface Model over North America, using an Ensemble Kalman Filter. The additional skill generated by each assimilation experiment is primarily evaluated using four years of in situ near-surface and root-zone soil moisture observations from 91 Soil Climate Analysis Network (SCAN) sites. These sites are distributed throughout the contiguous United States, sampling each of the main land cover types present. It is first confirmed that separately assimilating either of the X-band LPRM AMSR-E soil moisture or the ASCAT Surface Degree of Saturation data sets generates a significant improvement in the mean model soil moisture skill across all of the SCAN sites. It is then demonstrated that assimilating both of these data sets together further improved the mean model soil moisture skill over the SCAN sites. Additionally, when the assimilation results are considered separately for each land cover type, there is some contrast between the skill gained from assimilating either the passive or active data, while in nearly all cases the best results were obtained by assimilating both data types together.