Although many studies have documented developmental change in mathematics motivation, little is known about how these trends predict math performance. A sample of 288 participants from the United States reported their perceived math ability, math utility value and math interest in 5th, 7th and 9th grades. Latent growth curve models estimated developmental trajectories in each of these constructs. Mathematics interest and utility value decreased across time, but there was no significant change in self-perceived math ability. Slopes and intercepts of all mathematics motivation variables correlated with one another. Even when controlling for prior mathematics performance, students who self-reported high math ability in 5th grade had higher standardised test scores than their peers in high school five years later. Neither math utility value nor math interest intercepts or slopes predicted later performance. Understanding the predictors of math performance is important for supporting students' success in science, technology, engineering and mathematics careers.