Understanding treatment effect heterogeneity has been an increasingly important task in variousfields. Treatment effect heterogeneity not only adds granularity to the understanding of everyday matters but also assists better-informed decision-making on many scientific frontiers. In biomedical studies, learning treatment effect heterogeneity helps clinicians to apply personalized treatments to patient subpopulations with different genetic profiles. Instead of prescribing one drug for all, refined prescription strategies can potentially improve patients’ overall welfare. In social science studies, evaluating the treatment effect heterogeneity of candidate policies provides guidance for policymakers to implement future social programs. In technology companies, understanding treatment effect heterogeneity helps decision-makers to depict market segregation so that advertisement budgets can be strategically allocated to particular consumer subpopulations among which a new product is more likely to earn profits. This dissertation provides a set of statistical methodologies for understanding the treatment effect heterogeneity and is organized into three chapters with three separate aims: (1) estimating treatment effect heterogeneity, (2) confirming treatment effect heterogeneity, and (3) designing adaptive experiments toward learning treatment effect heterogeneity Chapter 1 introduces a statistical methodology aiming to estimate treatment effect heterogeneity efficiently. We take a model-free semiparametric perspective and aim to efficiently evaluate the heterogeneous treatment effects of multiple subgroups simultaneously under the one-step targeted maximum-likelihood estimation framework. When the number of subgroups is large, we further expand this path of research by looking at a variation of the one-step TMLE that is robust to the presence of small estimated propensity scores in finite samples. Chapter 2 proposes a statistical methodology for confirming the estimated heterogeneous treatment effects. Understanding the impact of the most effective treatments on outcome variables is crucial in various disciplines. Due to the widespread winner’s curse phenomenon, conventional statistical inference assuming that the top policies are chosen independent of the random sample may lead to overly optimistic evaluations of the best policies. In addition, given the increased availability of large datasets, such an issue can be further complicated when researchers include many covariates to estimate the policy or treatment effects in an attempt to control for potential confounders. To simultaneously address the above-mentioned issues, we propose a resampling-based procedure that not only lifts the winner’s curse in evaluating the best policies observed in a random sample but also is robust to the presence of many covariates. The proposed inference procedure yields accurate point estimates and valid frequentist confidence intervals that achieve the exact nominal level as the sample size goes to infinity for multiple best policy effect sizes. Chapter 3 provides an alternative perspective of studying the treatment effect heterogeneity. While much of the existing work in this research area has focused on either analyzing observational data based on untestable causal assumptions or conducting post hoc analyses of existing randomized controlled trial data, little work has gone into designing randomized experiments specifically for uncovering treatment effect heterogeneity. In this chapter, we develop a unified adaptive experimental design framework towards better learning treatment effect heterogeneity by efficiently identifying subgroups with enhanced treatment effects from a frequentist viewpoint. The adaptive nature of our framework allows practitioners to sequentially allocate experimental efforts adapting to the accrued evidence during the experiment. The resulting design framework can not only complement A/B tests in e-commerce but also unify enrichment designs and response adaptive randomization designs in clinical settings. Our theoretical investigations illustrate the trade-offs between complete randomization and our adaptive experimental algorithms.