This essay discusses principles of confounder selection in causal inference. Selecting appropriate confounders is critical for reliable causal analysis. When complete knowledge of a causal diagram is available, graphical rules can guide confounder selection. However, such knowledge is often unavailable. The paper proposes a practical approach for confounder selection when knowledge of each covariate's role (as a cause of exposure or outcome) is available. It suggests controlling for all variables that are causes of the exposure, outcome, or both, excluding instrumental variables, and including proxies for unmeasured common causes of exposure and outcome. The paper also relates these principles to statistical methods for covariate selection. It highlights the limitations of statistical methods alone, emphasizing the need for substantive knowledge. The "disjunctive cause criterion" is proposed as a key principle, which controls for variables that are causes of the exposure or outcome. It also notes that controlling for instrumental variables can introduce bias and that proxies for unmeasured confounders may be useful. The paper discusses the importance of covariate timing and the challenges of confounder selection in observational studies. It also reviews statistical methods for confounder selection, including forward and backward selection, and the "change-in-estimate" approach. The paper concludes that a theoretically informed approach, combining the disjunctive cause criterion with appropriate statistical methods, is essential for reliable causal inference.This essay discusses principles of confounder selection in causal inference. Selecting appropriate confounders is critical for reliable causal analysis. When complete knowledge of a causal diagram is available, graphical rules can guide confounder selection. However, such knowledge is often unavailable. The paper proposes a practical approach for confounder selection when knowledge of each covariate's role (as a cause of exposure or outcome) is available. It suggests controlling for all variables that are causes of the exposure, outcome, or both, excluding instrumental variables, and including proxies for unmeasured common causes of exposure and outcome. The paper also relates these principles to statistical methods for covariate selection. It highlights the limitations of statistical methods alone, emphasizing the need for substantive knowledge. The "disjunctive cause criterion" is proposed as a key principle, which controls for variables that are causes of the exposure or outcome. It also notes that controlling for instrumental variables can introduce bias and that proxies for unmeasured confounders may be useful. The paper discusses the importance of covariate timing and the challenges of confounder selection in observational studies. It also reviews statistical methods for confounder selection, including forward and backward selection, and the "change-in-estimate" approach. The paper concludes that a theoretically informed approach, combining the disjunctive cause criterion with appropriate statistical methods, is essential for reliable causal inference.