This paper provides a theory of earnings quality metrics, explaining what they capture and how they are related with each other. We develop a simple economy in which a manager has market price, earnings, and smoothing incentives and can bias earnings reports. This economy provides a framework for the analysis of earnings quality metrics. We express value relevance, persistence, predictability, smoothness, and discretionary accruals in terms of their primitives, the manager’s incentives, operating risk, and accounting noise, and examine their behavior on varying these primitives. We find that some of them, particularly value relevance and persistence, trace closely the change in information content in our setting. In contrast, smoothing and discretionary accruals reflect the rational market reaction to the earnings reports inconsistently. These results provide guidance for understanding and selecting metrics in empirical tests.