Though it is generally believed that doing homework promotes better learning, no consensus has been achieved concerning its importance and relevance to students' achievement. The historical up-and-down status of public attitudes toward homework indicates that understanding about the role of homework in students' learning is far from comprehensive and clear. The literature shows that much research effort has been devoted to the relationship between amount of homework and achievement, but little on quality of homework and how it has been used. Applying a 2-level hierarchical linear model on the TIMSS 2003 data, this study investigated the relationship between classroom practices on homework and 8th grade students' mathematics achievement in Hong Kong in three dimensions: frequency and amount of homework, types of homework, and usage modes of homework. The results showed that while the time spent on daily mathematics homework had significantly positive effect on students' TIMSS results, no effect was observed on the frequency of homework assigned. Out of three types of homework, only homework of the problem/question type demonstrated significant effect. The practice of having students start homework in class had negative influence on students' learning outcome. Implications and suggestions for educational practice and future research were then discussed.