Health impact assessments for fine particulate matter (PM2.5) often rely on simulated concentrations generated from air quality models. However, at the global level, these models often run at coarse resolutions, resulting in underestimates of peak concentrations in populated areas. This study aims to quantitatively examine the influence of model resolution on the estimates of mortality attributable to PM2.5 and its species in the USA. We use GEOS-Chem, a global 3-D model of atmospheric composition, to simulate the 2008 annual average concentrations of PM2.5 and its six species over North America. The model was run at a fine resolution of 0.5 × 0.66° and a coarse resolution of 2 × 2.5°, and mortality was calculated using output at the two resolutions. Using the fine-modeled concentrations, we estimate that 142,000 PM2.5-related deaths occurred in the USA in 2008, and the coarse resolution produces a national mortality estimate that is 8 % lower than the fine-model estimate. Our spatial analysis of mortality shows that coarse resolutions tend to substantially underestimate mortality in large urban centers. We also re-grid the fine-modeled concentrations to several coarser resolutions and repeat mortality calculation at these resolutions. We found that model resolution tends to have the greatest influence on mortality estimates associated with primary species and the least impact on dust-related mortality. Our findings provide evidence of possible biases in quantitative PM2.5 health impact assessments in applications of global atmospheric models at coarse spatial resolutions.