This study evaluates kilometer-scale simulations of atmospheric river (AR) precipitation produced by the Simple Cloud Resolving E3SM Atmosphere Model (SCREAM). Four AR events that made landfall in the Puget Sound basin of the U.S. Pacific Northwest (PNW) are simulated using SCREAM with regional refinement mesh (RRM) at 3.125 km over the PNW region within a global domain at 25 km resolution. All four AR events are also simulated using the Weather Research and Forecasting (WRF) model in a regional domain comparable to the RRM over the PNW at 3 km resolution. An ensemble of WRF simulations is performed using different physics options and land surface models. For SCREAM, two versions of the model that differ primarily in the representation of the ice-cloud fraction scheme are used. While SCREAMv0 uses a relative humidity-based ice-cloud fraction, SCREAMv0R sets ice cloud fraction based on the cell-averaged ice mass mixing ratio. For all four AR cases, SCREAMv0R simulates stronger lower-level wind, precipitable water, and column-integrated moisture flux than SCREAMv0 and is closer to the ERA5 reanalysis. Accompanying the more realistic large-scale circulation, precipitation is enhanced over the Cascades Range and reduced over the Coastal Range in SCREAMv0R compared to SCREAMv0, resulting in a more realistic spatial distribution of precipitation as depicted by multiple observational precipitation products. The increased precipitation over the Cascade Range in SCREAMv0R can be attributed to the larger ice and mixed-phase mass mixing ratios, which increase the middle-level latent heating that increases the moisture convergence at the lower level. With a single physics option, SCREAMv0R performs similarly to the best-performing WRF simulations for all four AR events, showing promise for modeling extreme precipitation produced by ARs in regions with complex terrain for modeling AR-induced flooding.