This study determined the working memory (WM) components (executive, phonological short-term memory [STM], and visual-spatial sketchpad) that best predicted mathematical word problem-solving accuracy in elementary schoolchildren (N = 392). The battery of tests administered to assess mediators between WM and problem-solving included measures of domain-general knowledge (fluid intelligence, reading, calculation), domain-general processes (naming speed, inhibition), domain-specific knowledge (word-problem representation, planning) and domain-specific processes (estimation and numerical magnitude judgments). Structural equation modeling analyses indicated (a) the executive component of WM and phonological STM yielded significant direct paths to problem-solving accuracy in the fully mediated model; (b) domain general knowledge (reading, calculation), domain-specific knowledge (problem representation, magnitude judgments) and domain general processes (inhibition) uniquely mediated the relationship between components of WM (executive, phonological STM) and problem-solving accuracy; and (c) those processes failing to uniquely mediate the relationship between WM and problem-solving accuracy in the full mediation model were measures of fluid intelligence, planning and estimation. The results support the notion that both the executive and phonological storage components of WM draw upon some of the same resources in predicting problem-solving, but none of the aforementioned mediation variables completely compensated for the influence of these 2 WM components on children's mathematical word problem-solving. More importantly, the findings suggest there are multiple pathways that mediate the relationship between components of WM performance and problem-solving accuracy.