Canadian students enrolled in either French-immersion or English-instruction programs were followed from grades 2 to 3 (Mage = 7.8 years to 8.9 years; N = 244; 55% girls). In each grade, students completed two mathematical tasks that required oral language processing (i.e., word-problem solving and number transcoding from dictation) and two that did not (i.e., arithmetic fluency and number line estimation). Students in both English-instruction (n = 92) and French-immersion programs (n = 152) completed tasks in English. Students in French-immersion programs also completed word-problem solving and transcoding tasks in French. The models were framed within the Pathways to Mathematics model, with a focus on the linguistic pathways for students in English-instruction and French-immersion programs. For tasks with oral language processing, performance in grade 3 was predicted by students’ English receptive vocabulary for both English-instruction and French-immersion students, even when French-immersion students were tested in French, controlling for performance in grade 2. In contrast, for tasks without oral language processing, receptive vocabulary in either English or French did not predict performance in grade 3, controlling for performance in grade 2. These results have implications for teaching mathematics within the context of immersion education.