Identifying which early mathematics skills have the largest effects on later mathematics achievement has important implications. However, regression-based estimates often rely on untested assumptions: (a) Scores on different mathematics skills reflect unique constructs, and (b) other factors affecting early and later mathematics achievement are fully controlled. We illustrate a process to test these assumptions with a sample of third and fourth graders who completed measures of mathematics skills, working memory and motivation, and standardized mathematics and English language arts tests. Factor analyses indicated that mathematics skills largely reflect the same underlying construct. The skills that loaded highest on the general factor most predicted both later mathematics and English language arts, even after adjusting for working memory and motivation. Findings suggest that relations between earlier mathematics and later achievement largely reflected more general factors that contribute to children's learning. We discuss the importance of establishing construct validity in correlational studies.