In 2001, California instituted a statewide test measuring English proficiency for English learners, students who are not proficient in English. In 2003 and 2004, nearly 500,000 English learners in grades 1-5 took this test each year. The relationship between bilingual education receipt and English proficiency is estimated using value-added regression models for each section of the test--listening and speaking, reading, and writing. In these regression models, students in bilingual education have substantially lower English proficiency of 0.3 standard deviations or more compared with other English learners in first and second grades. In contrast, the difference between bilingual education and other programs is usually less than 0.1 standard deviations for students in grades 3-5. These results hold for ordinary least squares, school fixed effects, and propensity score models.