Calculating the burden of disease attributed to air pollution requires accurate estimation of population level exposures to pollutants. Although coverage of ground monitoring networks is increasing, these data are insufficient to independently estimate exposures globally. Information from other sources, such as satellite retrievals, chemical transport models and land use covariates must therefore be used in combination with ground monitoring data. Each of these data sources will have their own biases and uncertainties that may vary over space. Set within a Bayesian hierarchical modelling framework, the recently developed Data Integration Model for Air Quality (DIMAQ) integrates data from multiple sources and allows spatially-varying relationships between ground measurements and other factors that estimate fine particulate matter (PM2.5) concentrations. The outputs of the model are estimated exposures that can be combined with population estimates to produce population-level distributions of exposures for each country. DIMAQ was used to estimate exposures of PM2.5, together with associated measures of uncertainty, on a high-resolution grid (~11 km × 11 km) covering the entire globe for use in the 2016 WHO report ‘Ambient air pollution: A global assessment of exposure and burden of disease’, and in the 2015 and 2016 updates of the Global Burden of Disease. For 2015, 92% of the world’s population lived in areas that exceeded the WHO 10 µg/m3 guideline. Fifty percent of the global population resided in areas with PM2.5 concentrations above the WHO Interim Target 1 (IT-1 of 35 µgm-3); 64% lived in areas exceeding IT-2 (25 µgm-3); and 81% lived in areas exceeding IT-3 (15 µgm-3). Nearly all (86%) of the most extreme concentrations (above 75 µgm-3) were experienced by populations in China, India, Pakistan, and Bangladesh.