Mendelian randomization (MR) is a method for causal inference that utilizes the natural experiment in genetic inheritance. The idea of MR can be traced back to the dawn of modern statistics and genetics a century ago, although the terminology was not well known until the beginning of this century. Owing to the explosion of genomic datasets, the popularity of MR has been growing exponentially since then. In this talk, I will provide a give of the historical ideas and recent progress in MR. I will talk about an ongoing work on “almost exact Mendelian randomization” that unifies a lot of the old and new insights.