Using original, representative survey data, we document that analytical, routine, and manual job tasks can be measured with high validity, vary substantially within and between occupations, are significantly related to workers' characteristics, and are robustly predictive of wage differences between occupations and among workers in the same occupation. We offer a conceptual framework that makes explicit the causal links between human capital endowments, occupational assignment, job tasks, and wages, which motivate a Roy model of the allocation of workers to occupations. We offer two simple tests of the model's gross predictions for the relationship between tasks and wages, both of which receive qualified empirical support.