RAN (Rapid Automatized Naming) is known to be a robust predictor of reading development in different languages. Much less is known about RAN predictive power in bilingual contexts. This is the first meta-analysis of research with bilingual children, assessing the strength of the RAN-reading relationship both within and across languages. It also explored the moderators that may affect this relationship. The search identified 38 published studies of bilingual children with 47 samples, 313 effect sizes and 5312 participants. Analyses of random-effects models with robust variance estimation revealed weak-to-moderate overall effect sizes of RAN and reading concurrently (r = -0.39) and longitudinally (r = -0.38). Moderator analyses of concurrent correlations revealed that RAN correlated more strongly with reading fluency (r = -0.56) than accuracy (r = -0.38). Alphanumeric RAN tasks (digits r = -0.39, letters r = -0.42) showed stronger associations with reading than non-alphanumeric RAN tasks (objects r = -0.38, colors r = -0.25). RAN-reading correlation was statistically significant both within and across languages. It was somewhat weaker when the two skills were measured in different languages (rL1RAN-L2 reading = -0.34, rL2RAN-L1 reading = -0.36) compared to when they were measured in the same language (rL1 = -0.40, rL2 = -0.44), though those differences failed to reach statistical significance. In addition, the type of bilingualism was found to be a potential moderator of the RAN-reading relationship longitudinally, with its magnitude being the strongest in simultaneous bilinguals. Our results suggest that, as a predictor, RAN taps into general, language-independent processes underlying reading.