This study investigated the relationship between teachers' classroom ratings of working memory (WM) and laboratory measures of WM of English language learners (ELLs) in the elementary grades. Multilevel modeling was used to identify whether teacher ratings accurately predicted ELL children's performance on WM tasks. The results indicated that teachers' ratings of WM were predicted by isolated components of WM even when measures of achievement, vocabulary and gender representation were entered into the analysis. Teacher ratings of WM were predictive of student performance on latent measures of Spanish short‐term memory and the executive component of WM. Entering inattention ratings into the analysis partialed out the influence of the executive component of WM, while leaving variance related to a language‐specific storage system as a significant predictor of teachers' classroom ratings of WM. The results suggest that classroom ratings of WM provide a valid analogue of student laboratory performance. Copyright © 2016 John Wiley & Sons, Ltd.