This nonexperimental study explored the predictive strength of English proficiency levels on academic achievement of middle school students in a sample of 17,470 native English-speaking (NES) students, 558 English language learners (current ELLs), and 500 redesignated fluent English proficient students (former ELLs). Results of multilevel analyses indicated that after controlling for relevant student' and school-level characteristics, former ELLs significantly outperformed current ELL and NES students in reading (effect sizes: 1.07 and 0.52) and mathematics (effect sizes: 0.86 and 0.42). The results support Cummins's (1979, 2000) lower level threshold hypothesis predicting that upon reaching adequate proficiency in the language of schooling and testing, ELLs would no longer experience academic disadvantages. Refinements for the theory and directions for future research are discussed. (Contains 3 tables and 13 notes.)