This study addresses the challenge of determining the causal direction between two traits when their measurements are imprecise, using genome-wide association study (GWAS) summary data. The authors demonstrate that traditional causal inference methods, such as the causal inference test (CIT), can incorrectly infer causality due to measurement error in phenotypes. They propose an extension of Mendelian randomization (MR) that allows for more reliable inference of causal direction, particularly in the presence of measurement error or unmeasured confounding. This method uses summary-level data from GWAS and is less susceptible to bias from measurement error. The authors apply this method to infer the causal direction between DNA methylation and gene expression levels. Their results show that DNA methylation is more likely to be the causal factor, but this conclusion is sensitive to systematic differences in measurement error between platforms and horizontal pleiotropy. The study emphasizes the importance of using MR alongside other methods like CIT to triangulate reliable causal conclusions. The authors also highlight the need for sensitivity analyses to assess the robustness of causal inferences under varying levels of measurement error. The study demonstrates that measurement error significantly affects causal inference, and that methods like MR can provide more reliable results in such scenarios. The findings underscore the importance of considering measurement error in causal inference studies and the value of using multiple methods to assess causal relationships.This study addresses the challenge of determining the causal direction between two traits when their measurements are imprecise, using genome-wide association study (GWAS) summary data. The authors demonstrate that traditional causal inference methods, such as the causal inference test (CIT), can incorrectly infer causality due to measurement error in phenotypes. They propose an extension of Mendelian randomization (MR) that allows for more reliable inference of causal direction, particularly in the presence of measurement error or unmeasured confounding. This method uses summary-level data from GWAS and is less susceptible to bias from measurement error. The authors apply this method to infer the causal direction between DNA methylation and gene expression levels. Their results show that DNA methylation is more likely to be the causal factor, but this conclusion is sensitive to systematic differences in measurement error between platforms and horizontal pleiotropy. The study emphasizes the importance of using MR alongside other methods like CIT to triangulate reliable causal conclusions. The authors also highlight the need for sensitivity analyses to assess the robustness of causal inferences under varying levels of measurement error. The study demonstrates that measurement error significantly affects causal inference, and that methods like MR can provide more reliable results in such scenarios. The findings underscore the importance of considering measurement error in causal inference studies and the value of using multiple methods to assess causal relationships.