Health effects from air pollution are a top ten cause of disability worldwide. Although research has linked air pollution to health, critical questions remain. This work examines key aspects of air pollution's impact on human health, including: (1) use of regional air quality modeling to estimate air pollution exposure in the United States (U.S.) for use in health studies; (2) relationships between air pollution and hospital admissions in the U.S., including estimates for populations and locations that cannot be studied using monitoring data alone; and (3) factors affecting vulnerability to health effects from air pollution exposure in São Paulo, Brazil. Most epidemiological studies use ambient monitors to estimate air pollution exposure, and while there is a growing body of research in industrializing nations, studies tend to be concentrated in North America and Europe. Use of ambient monitoring data is an efficient approach but restricts studies to locations and populations near monitors. The first project investigates the use of an emerging method, regional air quality modeling, to estimate air pollutant exposure in the U.S. The second project investigates the hypothesis that relationships between air pollution and health are not fully captured in studies relying exclusively on monitoring data, by using modeled and measured exposures to estimate health effects for populations without monitoring data. The third project investigates vulnerability to air pollution-related mortality in São Paulo, Brazil by sex, age, education, race, marital status, and neighborhood-level socioeconomic status (SES). I also examine whether the relationship between air pollution and mortality differs depending on the location at time of death, e.g., in-hospital versus out of hospital. Results from the first project demonstrate that exposure estimates derived from regional air quality modeling covered 50% more population compared to monitor-derived estimates. Counties with monitors were more urban, with a higher percentage of college graduates, and lower percentage of individuals living in poverty. Modeled estimates provided higher spatial and temporal resolutions, but model performance varied by pollutant, season, and location. Air quality models allow estimation of air pollution exposure for populations that differ demographically from those near monitors, but model output also have temporal and spatial biases that should be considered. The second project utilizes air quality modeling to estimate exposures and health impacts in areas and times with and without monitoring data, evaluate possible effect modification by community-specific characteristics (e.g., urbanicity, income), and investigate health effect estimates associated with short-term cumulative exposures. Higher cardiovascular effect estimates were observed in more urban counties. Compared to single day lags, higher respiratory health effect estimates were observed for multi-day lags measuring short-term cumulative exposures. Lag structures differed for respiratory and cardiovascular hospital admissions: the highest effect estimate for cardiovascular hospitalizations was observed at lag 01 (0.89% [PI: 0.51%, 1.28%] per 10µg/m3 PM2.5 [particulate matter with an aerodynamic diameter ≤2.5μ]), while the highest effect estimate for respiratory hospitalizations was observed at lag 06 (2.47% [0.29%, 4.69%] per 10µg/m3 PM2.5 ). Timing and pattern of exposure and health impacts may differ for respiratory and cardiovascular outcomes, and respiratory effect estimates based on a single day of exposure could underestimate the true effect. The third project contributes to knowledge of air pollution and health in São Paulo by investigating the role of individual- and community-level characteristics and the relationship between air pollution and mortality. Higher effect estimates for non-accidental mortality were observed for those with lower education: effects were 1.66% (0.23%, 3.08%), 1.51% (0.51%, 2.51%), and 2.82% (0.23%, 5.35%) higher for NO2, SO2, and CO exposure, respectively, in those with no education compared to those with postsecondary education. For cardiovascular mortality and PM10 exposure, effect estimates were 3.74% [0.044%, 7.30%] higher for no education compared to those with postsecondary education. Educational attainment is correlated with SES, but consistent trends of increasing health risks associated with pollutant exposure and residential SES were not observed. Positive, statistically significant associations between all air pollutants and mortality were observed for in-hospital deaths, whereas non-hospital deaths generally exhibited positive but not statistically significant associations. Scientific understanding of how air pollution affects health, including potentially vulnerable populations, requires new methods and analysis to answer questions that cannot be addressed with traditional approaches. Benefits from this research include the ability to estimate exposures for populations in times and locations without ambient monitors. This work provides some of the first estimates of air pollution-health risk using ambient modeling, estimates of risk for understudied rural populations, and potentially improved estimates for urban settings. The findings also provide evidence regarding which populations face the highest health burdens from air pollution.