This paper reviews matching methods for causal inference and outlines future directions. Matching methods aim to replicate randomized experiments by creating treated and control groups with similar covariate distributions, reducing bias. These methods are increasingly used in economics, epidemiology, medicine, and political science, but the literature is scattered across disciplines. The paper synthesizes existing research, providing a structured overview of matching methods and guidance on their use. Matching broadly refers to methods that balance covariates between treated and control groups. It is used in nonexperimental studies, which require careful design due to the absence of randomization. Matching is a key tool in the design stage of such studies, helping to ensure that treated and control groups are comparable before analyzing outcomes. While matching is primarily used for causal inference, it can also address noncausal questions, such as racial disparities. Alternatives to matching include regression adjustment, instrumental variables, structural equation modeling, and selection models. Matching has several advantages: it is complementary to regression adjustment, highlights areas of covariate distribution with insufficient overlap, and provides straightforward diagnostics for assessing performance. The paper outlines four key steps in implementing matching methods: defining closeness (distance measure), performing the matching, diagnosing the matched samples, and estimating the treatment effect. It discusses various distance measures, including exact matching, Mahalanobis distance, propensity score, and linear propensity score. Propensity scores summarize covariates into a single scalar, facilitating balance and reducing bias. Matching methods include nearest neighbor matching, optimal matching, ratio matching, and full matching. These methods vary in how they select matches and handle control individuals. Subclassification, full matching, and weighting are also discussed, with weighting methods such as inverse probability of treatment weighting (IPTW) and kernel weighting used to adjust for confounding. A key issue in matching is common support, which refers to the overlap in propensity score distributions between treated and control groups. Methods like subclassification and weighting may use all individuals, but it is important to restrict analysis to those within common support to ensure reliable estimates. The paper concludes with suggestions for future research, emphasizing the need for more systematic evaluation of propensity score estimation and the importance of accurate propensity score models in weighting and matching methods.This paper reviews matching methods for causal inference and outlines future directions. Matching methods aim to replicate randomized experiments by creating treated and control groups with similar covariate distributions, reducing bias. These methods are increasingly used in economics, epidemiology, medicine, and political science, but the literature is scattered across disciplines. The paper synthesizes existing research, providing a structured overview of matching methods and guidance on their use. Matching broadly refers to methods that balance covariates between treated and control groups. It is used in nonexperimental studies, which require careful design due to the absence of randomization. Matching is a key tool in the design stage of such studies, helping to ensure that treated and control groups are comparable before analyzing outcomes. While matching is primarily used for causal inference, it can also address noncausal questions, such as racial disparities. Alternatives to matching include regression adjustment, instrumental variables, structural equation modeling, and selection models. Matching has several advantages: it is complementary to regression adjustment, highlights areas of covariate distribution with insufficient overlap, and provides straightforward diagnostics for assessing performance. The paper outlines four key steps in implementing matching methods: defining closeness (distance measure), performing the matching, diagnosing the matched samples, and estimating the treatment effect. It discusses various distance measures, including exact matching, Mahalanobis distance, propensity score, and linear propensity score. Propensity scores summarize covariates into a single scalar, facilitating balance and reducing bias. Matching methods include nearest neighbor matching, optimal matching, ratio matching, and full matching. These methods vary in how they select matches and handle control individuals. Subclassification, full matching, and weighting are also discussed, with weighting methods such as inverse probability of treatment weighting (IPTW) and kernel weighting used to adjust for confounding. A key issue in matching is common support, which refers to the overlap in propensity score distributions between treated and control groups. Methods like subclassification and weighting may use all individuals, but it is important to restrict analysis to those within common support to ensure reliable estimates. The paper concludes with suggestions for future research, emphasizing the need for more systematic evaluation of propensity score estimation and the importance of accurate propensity score models in weighting and matching methods.