We explore the impact of the language of instruction on academic performance of Arabic-speaking students who are expected to learn mathematics and science in and through a second language (L2). Using as a theoretical background Cummins's (1984) framework relating language proficiency to academic achievement, we provide some insights into the relationship between native language, second language proficiency, and academic achievement in mathematics and science among students whose "mother tongue" is Arabic. We employ a propensity score technique and use TIMSS 2019 standardized tests to explain the differences in achievement in mathematics and science between two groups of young children living in an Arabic dialect-dominated environment who get instruction in English and who receive instruction in classical Arabic. Our findings highlight that the language of instruction accounts for the variations in performance.