This meta-analysis synthesizes the last two decades of experimental and quasi-experimental research on reading instruction across academic contexts (e.g., social studies, science, mathematics, English language arts) for English learners (ELs) in grades 4 through 8, to determine (a) the overall effectiveness of reading instruction for upper elementary and middle school students who are ELs and (b) how the magnitude of the effect varies based on student, instructional, and study characteristics. The analysis included a total of 11 studies with 46 individual effect sizes and yielded a mean effect size of g = 0.35 across all (i.e., standardized and unstandardized) reading measures, g = 0.01 across standardized reading measures, and g = 0.43 across unstandardized reading measures. For all reading, unstandardized reading, all vocabulary, and unstandardized vocabulary measures, results suggest that higher quality studies tended to have smaller effects, and these effects were even more evident for unstandardized measures (i.e., one unit increase in study quality was associated with decreased effects: g = 0.21, g = 0.30, g = 0.24, g = 0.30, respectively). For all comprehension measures, effects were larger for instruction that included both vocabulary and comprehension (g = 0.39) than for instruction that focused on vocabulary alone (g = 0.08). Results suggest the benefit of developing and refining high-impact approaches to reading instruction for ELs that can be delivered across content areas and grades.