The standard approach for assimilating satellite radiance observations is to interpolate all vertical levels of the background state and analysis increment to the same horizontal location for input to the radiative transfer model. This can add significant error for observations with large zenith angle. The impact of accounting for the true slanted satellite-viewing geometry was tested by modifying the horizontal interpolation routines in Environment and Climate Change Canada’s Global Deterministic Weather Prediction System. Consequently, model variables are interpolated to a different horizontal position at each model level, for either just the innovation or both the innovation and increment calculation. When this slant-path operator is used for simulation of radiances, reductions in innovation standard deviation, up to 4.5%, for upper-tropospheric and stratospheric temperature and humidity channels of ATMS, AMSU-A, MHS, and CrIS instruments have similar magnitudes as reported in previous studies. In data assimilation experiments, statistically significant reductions in innovation standard deviation (up to 0.3%) for global GPSRO observations are obtained, due to an improved background state. Verification of short- and medium-range forecasts against Era5, and own analyses over the region poleward of 60°S show statistically significant reductions of error standard deviation by 2%-3% for wind and temperature in upper troposphere and lower stratosphere. These positive impacts are mostly due to performing slant-path interpolation on the background state, while also using slant-path on the analysis increment has little additional impact. This is expected since the analysis increment in this global configuration has lower spatial resolution, with grid spacing comparable with the maximum horizontal position error from not using the slant-path operator.