Randomization is a fundamental principle in causal inference and was first proposed by R A Fisher about a century ago. Although randomization has now been universally adopted in the design of experiments, its role in the analysis of experiments and in observational studies remains controversial. This is partly due to a lack of precise description and understanding of the randomization principle. This talk will try to use modern tools in causal inference to better understand randomization and will have two parts. In the first part, I will use the potential outcomes framework to define the conditional randomization test, distinguish it with related ideas, and discuss some recent applications. In the second part, I will use the causal graphical model framework to understand Mendelian randomization, a recently popularized natural experiment design that is closely linked to the inception of the randomization principle.