Students' intelligence and self-concept of ability are critical predictors of school achievement. However, studies focusing on the relative importance of both achievement predictors have produced mixed results. To clarify these inconsistencies, the present study investigated whether the relative predictive power of students' intelligence and ability self-concept differs depending on the achievement indicator at hand (standardized test performance vs. school grades) and the achievement domain (math vs. language arts). Data from 1,067 German 8th graders were analyzed using multilevel structural equation modeling. Our results showed that students' intelligence was the best unique predictor of standardized test performance in math, but its predictive power for students' test performance in German was comparable with the predictive effects of students' ability self-concept in this domain. Students' domain-specific ability self-concept emerged as the best predictor of students' grades in both math and German. Commonality analyses revealed greater overlap between students' intelligence and self-concept in math than in the verbal domain. Finally, analyses of compositional effects indicated that having classmates with comparatively high intelligence positively predicted students' standardized test performance in both math and German, controlling for individual intelligence, but did not significantly predict students' grades. Students' gender, socioeconomic status (SES), immigration background, and academic track level were included as covariates in these analyses. The present study underscores the importance of considering different achievement indicators in analyses of the cognitive and motivational underpinnings of student achievement across academic domains.