Including English language learners (ELLs) in large-scale assessments raises questions about the validity of inferences based on their scores. Test accommodations for ELLs are intended to reduce the impact of limited English proficiency on the assessment of the target construct, most often mathematic or science proficiency. This meta-analysis synthesizes research on the effectiveness and validity of such accommodations for ELLs. Findings indicate that none of the seven accommodations studied threaten the validity of inferences. However, only one accommodation--providing English dictionaries or glossaries--has a statistically significant effect on ELLs' performance, and this effect equates to only a small reduction in the achievement score gap between ELLs and native English speakers. Findings suggest that accommodations to reduce the impact of limited language proficiency on academic skill assessment are not particularly effective. Given this, we posit a hypothesis about the necessary role of academic language skills in mathematics and science assessments. (Contains 5 notes and 2 tables.)