English learners (EL) are one the largest population groups within the public schools in United States. Their academic achievement is lower than their monolingual peers in several areas. Working memory (WM), a cognitive process, has been identified as one of the predictors of poor academic achievement in both monolingual English speaking and EL children. Children at risk for learning problems have poor WM capacity. In addition, teachers have been acknowledged as one of the primary sources for identifying children at risk for learning problems. This study investigated the relationship between teachers’ ratings of ELs’ WM and students’ actual WM performance. Because WM is composed of several components, a determination was made as to whether teacher ratings are related to isolated components of WM or whether ratings are predictive of all components. Because teacher ratings are context specific (student make-up of classroom), multilevel modeling was used to identify whether teacher ratings accurately predict EL children’s performance on WM tasks. Additionally, covariates related to student characteristics and achievement were entered in the model to determine whether unique variance related to teacher ratings was sustained in predictions of WM performance. The results suggest that teacher ratings do predict both general WM and isolated components of WM in both English and Spanish. However, this relationship was not sustained when covariates related to student achievement were entered into the multilevel regression model. The results suggest that teacher ratings of low WM were more related to vocabulary and academic achievement than WM limitations.