Previous research provides conflicting recommendations on how to interpret the interaction effect between a focal predictor and a mediator. There exists a debate on whether it is possible to examine dual (mediating and moderating) effects of the mediator with the interaction effect. Furthermore, there exists a debate on whether one should interpret the interaction effect as an indirect effect or a direct effect. Another stream of research recommends one not to consider the interaction effect, because it is not possible to interpret the interaction effect clearly. This paper shows that the interaction effect can be decomposed into two components in view of the marginal effects of a focal predictor: a direct effect and an indirect effect. Based on the results, this paper concludes that the interaction effect captures the dual effects of a third variable as a moderator as well as a mediator. Then, it recommends one to interpret the interaction effect as the sum of a direct effect and an indirect effect. In addition, it provides a way of classifying mediators into three types according to their roles.