Children who go on to develop reading difficulties are a remarkably heterogeneous group, and accurately predicting who will need intervention before a child falls behind is not possible with present behavioral measures. A few key behavioral measures, such as rapid automatized naming (RAN), which assesses the speed of naming an array of familiar items, and phonological awareness (PA) have emerged as the strongest correlates of future reading performance. RAN predicts reading across different ages, ability levels, orthographies, and even across time. In this systematic review and meta-analysis (with N = 68 samples; k = 373 effect sizes; n = 10513 participants), we test the extent to which measures of RAN assessed before grade school predict later reading performance. We also test whether characteristics of the RAN tasks, reading measures, or sample demographics moderate this relationship. We found that kindergarten/preschool RAN is correlated with grade-school reading at r = -.38, similar in magnitude to previous meta-analyses that included various ages and languages. We found that alphanumeric RAN is particularly strongly related to future reading (p = .01) but that other features of the RAN task do not alter its predictive significance. RAN predicts reading for all kinds of measures, but more strongly for real words than nonwords (p < .001). These results support a shared cognitive resource model in which the similarity between RAN and reading tasks accounts for their correlation. We provide practical guidelines based on these data for early screening for reading difficulties and dyslexia.