Second language (L2) vocabulary learning has been deemed a daunting task for many students. This meta-analysis study aimed to explore the effectiveness of applying digital games for L2 vocabulary learning. A total of 26 published studies (2001–2017) conformed with the inclusion/exclusion criteria. Due to diverse findings of previous meta-analysis research in the field, we propose a framework of four-condition research designs to differentiate the empirical studies in an attempt to disclose possible expositions for the diversity and to connect the specific learning mechanisms with the research evidence. The overall effect sizes of the studies in the four conditions are reported as follows: A large overall effect size for Condition 1 (10 studies) (experimental groups playing digital games versus control groups receiving alternative activities), medium for Condition 2 (experimental groups playing digital games with a feature added or changed versus control groups playing base-version games) (10 studies), medium to large for Condition 3 (experimental groups playing digital games and control/comparison groups receiving identical content via conventional means) (two studies), and non-significant for Condition 4 (all participants playing the same digital games but being grouped by a non-game related variable) (four studies). Next, a structure diagram is developed in which the four conditions of the research design are connected with their respective game-related factors based on their locality. Further, we conducted moderator analyses to examine how the eight potential moderator variables (game design, educational level, L2 proficiency level, linguistic distance, intervention setting, assessment type, game source and intervention length) influenced the effect sizes in Conditions 1 and 2 to illustrate various digital game-based L2 vocabulary learning scenarios. Finally, suggestions and implications are provided for game designers, educational practitioners, and researchers in the field.