The present study tested the "interference hypothesis"-that learning and using more advanced representations and strategies requires the inhibition of prior, less advanced ones. Specifically, it examined the relation between inhibitory control and number line estimation performance. Experiment 1 compared the accuracy of adults' (N = 53) estimates on 2 number line tasks, 1 with standard (power of 10) endpoints (0-1,000) and the other with nonstandard endpoints (364-1,364). Inhibition, as measured by Stroop task performance, predicted the accuracy of estimation on the nonstandard number line task, above and beyond estimation on the standard task. In Experiment 2, changes in kindergartners' (N = 42) 0-100 number line estimation were elicited through randomized training conditions, which involved playing a numerical board game. Stroop task performance was related to the rate of improvement in estimation, controlling for pretest number line task performance. The results provide a potential explanation for the relation between inhibitory control and mathematics achievement: Individuals with better inhibitory control may be better able to suppress the activation of prior knowledge and may be less vulnerable to interference from such knowledge. Potential implications for instructional design are discussed.