The hypothesis that native language (L1) proficiency promotes English acquisition and overall academic achievement, a key theoretical assumption underlying bilingual education, was tested using latent growth modeling of data from 899 limited-English-proficient (LEP) eighth graders who were followed for 12 years in the National Education Longitudinal Study (NELS:88/2000). A model in which L1 proficiency predicted English (L2) reading ability, which in turn predicted high school achievement and distal educational/occupational attainment, fit the data well for the full LEP sample and a Hispanic subsample. In Hispanics, the model explained 24.1%, 7.4%, 29.4%, and 46.3% of the variance in initial English reading level, English reading growth, high school achievement, and post-high school attainment, respectively. Model fit for an Asian subsample, however, was poor. Possible reasons for lack of group invariance include cultural differences in construct conceptualization, greater linguistic and cultural heterogeneity within the Asian subgroup, and cross-language transfer difficulties when L1 and L2 lack a shared alphabetic structure. At least for Hispanic LEP students, this study's results establish the theoretical foundation for exploring the effectiveness of specific educational interventions. (Contains 3 tables, 6 figures and 10 footnotes.)