Interpersonal interaction is closely related to subjective well-being (SWB), but with the rise of social network sites (e.g., Facebook (FB)), a new form of interpersonal interaction has emerged. Taking FB as an example, this study examined the effects of realistic (face-to-face) and online (FB) interaction on SWB and tested the mediating effect of relationship quality. Previous research studies on the effect of Facebook use on SWB have revealed very inconsistent results (the so-called “Facebook paradox”). Our paper has exhaustively reviewed previous studies and proposed that FB interaction, defined as active and reciprocal interaction with friends through FB, as a type of Facebook use might be beneficial to SWB. Data were collected through an online questionnaire, and 412 participants (including students and non-students) filled in the scales of realistic interaction, online (FB) interaction, friendship quality, and SWB.  Among them, 38 cases who were not FB usersand/or showed response sets were excluded from further analysis. Using structural equation modeling, the data were analyzed among the valid sample (241 females, 136 males). The Sobel test and the bootstrapping method with 5000 bootstrap samples were used to test the significance of the mediating effects. The results showed that: (1) realistic interaction with friends has a significant positive effect on SWB, acting through the mediating variable friendship quality; (2) controlling for the effect of realistic interaction, online (FB) interaction with friends still has a significant positive effect on SWB, acting through the mediator friendship quality; and (3) both realistic and FB interaction have fairly equal effects on friendship quality, through a model comparison approach. We conclude that FB interaction is beneficial to SWB, beyond the effect of realistic interaction. Thus, our study provides one solution to the Facebook paradox and thus contributes to issues relating to FB interaction. Implications, limitations, and future directions are also discussed.