Streamflow predictions derived from a hydrologic model are subjected to many sources of errors, including uncertainties in meteorological inputs, representation of physical processes, and model parameters. To reduce the effects of these uncertainties and thus improve the accuracy of model prediction, the United States (U.S.) National Water Model (NWM) incorporates streamflow observations in the modeling framework and updates model-simulated values using the observed ones. This updating procedure is called streamflow data assimilation (DA). This study evaluates the prediction performance of streamflow DA realized in the NWM. We implemented the model using WRF-Hydro® with the NWM modeling elements and assimilated 15-min streamflow data into the model, observed during 2016–2018 at 140 U.S. Geological Survey stream gauge stations in Iowa. In its current DA scheme, known as "nudging," the assimilation effect is propagated downstream only, which allows us to assess the performance of streamflow predictions generated at 70 downstream stations in the study domain. These 70 locations cover basins of a range of scales, thus enabling a multi-scale hydrologic evaluation by inspecting annual total volume, peak discharge magnitude and timing, and an overall performance indicator represented by the Kling–Gupta efficiency. The evaluation results show that DA improves the prediction skill significantly, compared to open-loop simulation, and the improvements increase with areal coverage of upstream assimilation points.