Background: Measures of population-level influenza severity are important for public health planning, but estimates are often based on case-fatality and case-hospitalization risks, which require multiple data sources, are prone to surveillance biases, and are typically unavailable in the early stages of an outbreak. In this study, we develop a severity index based on influenza age dynamics estimated from routine surveillance data that can be used in retrospective and early warning contexts. Methods and Findings: Our method relies on the observation that age-specific attack rates vary between seasons, so that key features of the age distribution of cases may be used as a marker of severity early in an epidemic. We illustrate our method using weekly outpatient medical claims of influenza-like illness (ILI) in the United States from the 2001 to 2009 and develop a novel population-level influenza severity index based on the relative risk of ILI among working-age adults to that among school-aged children. We validate our ILI index against a benchmark that comprises traditional influenza severity indicators such as viral activity, hospitalizations and deaths using publicly available surveillance data. We find that severe influenza seasons have higher relative rates of ILI among adults than mild seasons. In reference to the benchmark, the ILI index is a robust indicator of severity during the period of peak epidemic growth (87.5% accuracy in retrospective classification), and may have predictive power during the period between Thanksgiving and the winter holidays (57.1% accuracy in early warning). We further apply our approach at the state-level to characterize regional severity patterns across seasons. We hypothesize that our index is a proxy for severity because working-age adults have both pre-existing immunity to influenza and a high number of contacts, infecting them preferentially in severe seasons associated with antigenic changes in circulating influenza viruses. Our analysis is limited by its application to seasonal influenza epidemics and a relatively short study period. Conclusions: Our severity index and research on the link between age dynamics and seasonal influenza severity will enable decision makers to better target public health strategies in severe seasons and improve our knowledge of influenza epidemiology and population impact. These findings demonstrate that routine surveillance data can be translated into operational information for policymakers. Our study also highlights the need for further research on the putative age-related mechanisms of severity in seasonal and pandemic influenza seasons.