This study investigated student and classroom level variables related to student science achievement by using the Trends in International Mathematics and Science Study (TIMSS) 2015 dataset. Multilevel modeling was applied in this secondary analysis to examine how science self-concept, school relatedness, and science teaching quality predicted student science achievement at both the student and classroom levels. Data from a total of 14,291 grade 8 students in the USA were included in the study. Results showed that after accounting for gender and SES (control variables), self-concept was a statistically significant, positive predictor of science achievement at the student level. At the classroom level, classroom-mean self-concept and sense of relatedness were both statistically significant, positive predictors of science achievement. Students' perceived teaching quality was not a statistically significant predictor at either the student or classroom level. Implications of these results for science teaching and learning are discussed.