It is often asserted that working memory predicts more variance in academic proficiency than do measures of intelligence. We used data from three studies to show that the validity of this assertion is highly dependent on the method of analysis. Using the same measures of intelligence, but different measures of working memory and algebraic proficiency, we found working memory provided better explanatory power only when analysis was conducted on the observed variable level. When the same data were analysed using structural equation models, only measures of intelligence had a direct effect on algebraic proficiency. From a theoretical viewpoint, our findings are consistent with a claim that working memory is a constituent component of (fluid) intelligence