Because discretionary accruals are known to be noisy proxies of earnings management that often produce biased results, we argue that a correlation between discretionary accruals and a hypothesized factor is generally not an adequate basis for valid inferences about earnings management. In this commentary, we propose and discuss a set of suggestions for improving the research design of studies on earnings management that use discretionary accruals. Specifically, we discuss suggestions for developing predictions about earnings management, choosing among discretionary accrual proxies, estimating discretionary accruals, addressing potential bias in the estimates, and critically evaluating the results obtained.