To the extent that pollution and population are spatially correlated, air quality modeling with coarse-resolution horizontal grids may systematically underpredict exposures and disparities in exposure among demographic groups (i.e., environmental injustice). We use InMAP, a reduced-complexity air pollution model, to quantify how estimates of year-2014 fine particulate matter (PM2.5) exposure in the United States vary with model spatial resolution, for a variable-resolution grid. We test five grids, with population-weighted average grid cell edge lengths ranging from 5.9 to 69 km. We find that model-estimated PM2.5 exposure, and exposure disparities among racial-ethnic groups, are lower with coarse grids than with fine grids: switching from our coarsest- to finest-resolution grid increases the calculated population-weighted average exposure by 27% (from 6.6 to 8.3 μg m-3) and causes the estimated difference in average exposure between minorities and whites to increase substantially (from 0.4 to 1.6 μg m-3). Across all grid resolutions, exposure disparities by race-ethnicity can be detected in every income category. Exposure disparities by income alone remain small relative to disparities by race-ethnicity, irrespective of resolution. These results demonstrate the importance of fine model spatial resolution for identifying and quantifying exposure disparity.