Using eighth-grade mathematics scores from TIMSS 2003, a large-scale international achievement assessment database, this study investigated correlates of math achievement in two developed countries, Canada and the United States and two developing countries, Egypt and South Africa. Variation in math achievement within and between schools for individual countries was accounted for by a series of two-level HLM models. Specifically, there were five sets of HLM models representing student background, home resources, instructional practices, teacher background, and school background related factors. In addition, a final model was built by including all the statistically significant predictors in earlier models to predict math achievement. Findings from this study suggested that whereas the instructional practices model worked the best for the United States and the teacher background model served as the most efficient and parsimonious model for predicting math achievement in Egypt, the final model served as the best model for predicting math achievement in Canada and South Africa. These findings provide empirical evidence that different models are needed to account for factors related to achievement in different countries. This study, therefore, highlights the importance that policy makers and educators from developing countries should not base their educational decisions and educational reform projects solely on research findings of developed countries. Rather, they need to use their country-specific findings to support their educational decisions. This study also provides a methodological framework for applied researchers to evaluate the effects of background and contextual factors on students’ math achievement.