Epidemiologic studies are crucial to the estimation of numbers of deaths attributable to air pollution. In this paper, the authors present a framework for distinguishing estimates of attributable cases based on time-series studies from those based on cohort studies, the latter being 5-10 times larger. The authors distinguish four categories of death associated with air pollution: A) air pollution increases both the risk of underlying diseases leading to frailty and the short term risk of death among the frail; B) air pollution increases the risk of chronic diseases leading to frailty but is unrelated to timing of death; C) air pollution is unrelated to risk of chronic diseases but short term exposure increases mortality among persons who are frail; and D) neither underlying chronic disease nor the event of death is related to air pollution exposure. Time-series approaches capture deaths from categories A and C, whereas cohort studies assess cases from categories A, B, and C. In addition, years of life lost can only be derived from cohort studies, where time to death is the outcome, while in time-series studies, death is a once-only event (no dimension in time). The authors conclude that time-series analyses underestimate cases of death attributable to air pollution and that assessment of the impact of air pollution on mortality should be based on cohort studies.