Mediation analysis is a critical tool in psychological research, helping to understand how or why variables are related by identifying mediating variables that transmit the effect of an antecedent variable on a dependent variable. This paper reviews current methods for investigating mediating variables, highlighting their importance in understanding underlying processes of human behavior. The single-mediator model is a fundamental framework in mediation analysis, where an antecedent variable (X) influences a mediator (M), which in turn influences a dependent variable (Y). The model includes equations that quantify direct, indirect, and total effects. The indirect effect is calculated as the product of the coefficients for the X-to-M and M-to-Y paths, or as the difference between the total and direct effects. Significance testing of the mediated effect involves various methods, including the causal-steps approach, which requires testing the significance of the overall X-to-Y relationship, the X-to-M relationship, and the M-to-Y relationship. However, this approach has limitations, such as low power in detecting mediated effects and the requirement for a significant overall effect. Confidence limits for the mediated effect are crucial for assessing the reliability of the estimate. Recent research suggests that normal distribution-based confidence limits may be inaccurate, and alternative methods, such as resampling techniques and product-based confidence limits, are more reliable. Longitudinal mediation models allow for the examination of stable mediated effects over time and provide insights into temporal precedence. These models include autoregressive, latent-growth, and latent-difference-score models, each with specific applications. Models that incorporate moderation and mediation are also discussed, highlighting the importance of understanding how mediated effects may vary across different levels of moderator variables. Experimental-design approaches to mediation combine statistical analysis with randomized experiments to test mediation theories more effectively. Other extensions of mediation analysis include models for binary dependent variables, multiple mediators, and multilevel data. Future directions in mediation analysis include longitudinal models, causal inference methods, and improved models for handling complex data structures. Overall, mediation analysis remains a vital area of research, offering insights into the mechanisms underlying variable relationships and informing more effective interventions.Mediation analysis is a critical tool in psychological research, helping to understand how or why variables are related by identifying mediating variables that transmit the effect of an antecedent variable on a dependent variable. This paper reviews current methods for investigating mediating variables, highlighting their importance in understanding underlying processes of human behavior. The single-mediator model is a fundamental framework in mediation analysis, where an antecedent variable (X) influences a mediator (M), which in turn influences a dependent variable (Y). The model includes equations that quantify direct, indirect, and total effects. The indirect effect is calculated as the product of the coefficients for the X-to-M and M-to-Y paths, or as the difference between the total and direct effects. Significance testing of the mediated effect involves various methods, including the causal-steps approach, which requires testing the significance of the overall X-to-Y relationship, the X-to-M relationship, and the M-to-Y relationship. However, this approach has limitations, such as low power in detecting mediated effects and the requirement for a significant overall effect. Confidence limits for the mediated effect are crucial for assessing the reliability of the estimate. Recent research suggests that normal distribution-based confidence limits may be inaccurate, and alternative methods, such as resampling techniques and product-based confidence limits, are more reliable. Longitudinal mediation models allow for the examination of stable mediated effects over time and provide insights into temporal precedence. These models include autoregressive, latent-growth, and latent-difference-score models, each with specific applications. Models that incorporate moderation and mediation are also discussed, highlighting the importance of understanding how mediated effects may vary across different levels of moderator variables. Experimental-design approaches to mediation combine statistical analysis with randomized experiments to test mediation theories more effectively. Other extensions of mediation analysis include models for binary dependent variables, multiple mediators, and multilevel data. Future directions in mediation analysis include longitudinal models, causal inference methods, and improved models for handling complex data structures. Overall, mediation analysis remains a vital area of research, offering insights into the mechanisms underlying variable relationships and informing more effective interventions.