Pleiotropy, where a single genetic variant influences multiple traits, is common in the human genome. Mendelian randomization (MR) uses genetic variants as proxies to estimate causal relationships between traits, assuming that apparent pleiotropy is due to vertical pleiotropy (the exposure influencing the outcome through the variant) rather than horizontal pleiotropy (independent pathways). Recent advances in genetic data have enabled more reliable MR analyses. This review discusses methods to detect and address horizontal pleiotropy, which can bias MR results. Key strategies include using summary data from genome-wide association studies (GWAS), colocalization methods to identify shared causal variants, and statistical techniques like MR-Egger regression to detect directional pleiotropy. Outlier removal methods, such as MR-PRESSO, help mitigate the impact of invalid instruments. Polygenic risk scores and multivariable MR approaches also enhance causal inference. Negative control analyses and model averaging further improve reliability. The review highlights the importance of combining multiple methods to address pleiotropy and improve the accuracy of causal estimates in MR studies.Pleiotropy, where a single genetic variant influences multiple traits, is common in the human genome. Mendelian randomization (MR) uses genetic variants as proxies to estimate causal relationships between traits, assuming that apparent pleiotropy is due to vertical pleiotropy (the exposure influencing the outcome through the variant) rather than horizontal pleiotropy (independent pathways). Recent advances in genetic data have enabled more reliable MR analyses. This review discusses methods to detect and address horizontal pleiotropy, which can bias MR results. Key strategies include using summary data from genome-wide association studies (GWAS), colocalization methods to identify shared causal variants, and statistical techniques like MR-Egger regression to detect directional pleiotropy. Outlier removal methods, such as MR-PRESSO, help mitigate the impact of invalid instruments. Polygenic risk scores and multivariable MR approaches also enhance causal inference. Negative control analyses and model averaging further improve reliability. The review highlights the importance of combining multiple methods to address pleiotropy and improve the accuracy of causal estimates in MR studies.