Whereas accuracy is used as an indicator of cognitive flexibility in preschool-age children, reaction time (RT), or a combination of accuracy and RT, provide better indices of performance as children transition to school. Theoretical models and cross-sectional studies suggest that a speed-accuracy tradeoff may be operating across this transition, but the lack of longitudinal studies makes this transition difficult to understand. The current study explored the longitudinal and bidirectional associations between accuracy and RT on the DCCS (mixed block) at 5, 6, and 7 years of age using cross-lagged panel analyses. The study also examined the roles of working memory and language, as potential longitudinal mediators between RT at Time X and accuracy at Time X + 1, and explored the role of inhibitory control. The sample consisted of 425 children from the Quebec Longitudinal Study of Child Development. Results show lagged associations from slower RT to greater improvements in accuracy between 5 and 6 years and between 6 and 7 years. Further, higher accuracy at 6 years predicted faster RT at 7 years. Only working memory acted as a partial mediator between RT at 5 years and accuracy at 6 years. These results provide needed longitudinal evidence to support theoretical claims that slower RT precedes improved accuracy in the development of cognitive flexibility, that working memory may be involved in the early stage of this process, and that accuracy and reaction time become more efficient in later stages of this process.