Several studies have shown that rapid automatized naming (RAN) is a significant predictor of mathematics, but the nature of their relationship remains elusive. Thus, the purpose of this meta-analysis was to estimate the size of their relationship and determine the conditions under which they correlate. We used a random-effects model analysis of data from 38 studies (33 unique samples, 151 correlations, 7,135 participants) to examine the size of the RAN–mathematics relationship and the role of different moderators (i.e., math measure and variable, type of RAN task, math age, study design, and sample characteristics). The results showed a significant correlation between RAN and mathematics (r = .37; 95% confidence interval [CI] [.33–.42]) as well as a large heterogeneity of individual correlations. The results also revealed that RAN produced stronger correlations with arithmetic calculation tasks than with general achievement tests; stronger correlations with single-digit calculation tasks than multidigit calculation tasks; and stronger correlations with math fluency tasks than math accuracy tasks. The effect of these moderators suggests that part of the reason why RAN predicts mathematics is that they both require quick access to and retrieval of phonological representations from long-term memory. Our findings also suggest that RAN objects or colors can be used as early predictors of mathematical skill, especially of arithmetic fluency.