This paper addresses the identification, inference, and sensitivity analysis of causal mediation effects. The authors prove that under a specific version of sequential ignorability, the average causal mediation effect (ACME) is nonparametrically identified. They compare this assumption with those in the literature and discuss its practical implications. The paper also shows that the popular estimator based on the linear structural equation model (LSEM) can be interpreted as an ACME estimator under additional parametric assumptions, which can be relaxed within and outside the LSEM framework. A new sensitivity analysis method is proposed to evaluate the robustness of empirical findings to unmeasured pre-treatment confounders. The methods are applied to a randomized experiment from political psychology, and easy-to-use software is provided.This paper addresses the identification, inference, and sensitivity analysis of causal mediation effects. The authors prove that under a specific version of sequential ignorability, the average causal mediation effect (ACME) is nonparametrically identified. They compare this assumption with those in the literature and discuss its practical implications. The paper also shows that the popular estimator based on the linear structural equation model (LSEM) can be interpreted as an ACME estimator under additional parametric assumptions, which can be relaxed within and outside the LSEM framework. A new sensitivity analysis method is proposed to evaluate the robustness of empirical findings to unmeasured pre-treatment confounders. The methods are applied to a randomized experiment from political psychology, and easy-to-use software is provided.