The present study examined the predictive validity of a classroom observation measure, the "Classroom Strategies Assessment System" (CSAS)-Observer Form, as a predictor of student performance on statewide tests of mathematics and English language arts. The CSAS is a teacher practice assessment that measures evidence-based instructional and behavioral management practices (Reddy and Dudek 2014). This study presents a sample of 35 teachers and 829 third through eighth grade students from six urban high poverty schools. Six school administrators trained to criterion on using the CSAS conducted three classroom observations for each teacher as part of their yearly evaluation process. Zero-order correlations revealed negative relationships between CSAS Rating Scale discrepancy scores (i.e., ? | recommended frequency-frequency ratings |) and mathematics and English language arts proficiency scores. A series of two-level hierarchical generalized linear models were fitted to data to assess whether CSAS Instructional Strategy and Behavioral Management Strategy Rating Scale Total and Composite discrepancy scores predicted statewide mathematics and English language arts proficiency scores. Results indicated that CSAS Total and Instructional and Behavior Management Composite discrepancy scores significantly predicted both mathematics and English language arts proficiency scores, suggesting that larger discrepancies on observer ratings of what teachers did versus what should have been done were associated with lower proficiency scores. Results offer evidence of the utility of measuring teacher practices via the CSAS to inform student achievement in high poverty settings. Implications for school psychological practice, teacher evaluation, and professional development are discussed.