In the cyber era, people interact with others not only face-to-face but also through social media platforms such as Facebook and Instagram. Social media addiction has emerged as a problem of global concern, with researchers all over the world conducting studies to evaluate how pervasive the problem is. However, the prevalence rates of social media addiction reported in the literature vary dramatically. The present meta-analysis aimed to systematically synthesize the extant research on social media addiction prevalence. Subgroup analysis and meta-regression were conducted to investigate whether the prevalence rates would differ by classification schemes, cultural values, and demographic factors. The meta-analysis involved 63 independent samples with 34,798 respondents from 32 nations spanning seven world regions. The random-effects meta-analytic findings revealed variations in prevalence among studies adopting distinct classification schemes. The pooled prevalence estimate was 5% (95% CI: 3%-7%) for studies adopting monothetic or strict monothetic classifications. A higher pooled prevalence estimate (13%; 95% CI: 8%-19%) was found for studies adopting a cutoff for severe level or strict polythetic classifications, and that estimate was even higher (25%; 95% CI: 21%-29%) for studies adopting a cutoff for moderate level or polythetic classifications. Moreover, cross-cultural comparisons revealed the pooled prevalence estimate obtained in collectivist nations (31%; 95% CI: 26%-36%) to be twofold higher than that obtained in individualist nations (14%; 95% CI: 9%-19%). This meta-analysis indicates that both the classification scheme used and cultural factors should be considered when interpreting the prevalence findings on social media addiction.