We explored the role of three math instructional practices by teachers (explicit instruction, guided practice, peer interaction) in predicting current Spanish and English reading comprehension and later Spanish and English math problem-solving performance of English learners (ELs). In this study, math was taught in Spanish. We had 26 teachers and 319 Spanish-speaking ELs from dual language classrooms. Of concern was whether the teachers' use of these three practices in the first year of the study yielded significant predictions in ELs' Spanish and English reading comprehension in the current year and math problem-solving accuracy in the subsequent years. Our multilevel modeling results indicated that math skills could predict reading performance and that early calculation in Spanish and English significantly predicted later problem-solving performance. The results also indicated a trade-off regarding the language of instruction and math ability. Further, such instructional practices were more likely to benefit ELs without learning difficulties in math (LDM) than those with LDM. These findings suggest that when teachers use the three instructional practices in early years, these practices need to account for the language of instruction and the continuum of student abilities related to math in order to better improve ELs' problem-solving accuracy in later years.