The purpose of this study was to investigate the effectiveness of explicit, direct, and generative strategy training and working memory capacity (WMC) on mathematical word problem-solving accuracy in elementary schoolchildren. In this study, children in third grade ("N" = 82) identified as at risk for math difficulties (MD) were randomly assigned (within classrooms) to one of three treatment conditions that explicitly directed students' attention to different propositions within word problems--paraphrase question propositions (Restate), paraphrase relevant propositions (Relevant), and paraphrase all propositions (Complete)--or an untreated control condition. A significant treatment by covariate design indicated that generative strategy outcomes were conditional on the level of pretest WMC. A clear advantage in posttest problem-solving accuracy and solution planning was found for the complete generative condition relative to the control condition, but this advantage was conditional on setting WMC to a high level. Although no significant treatment advantages were found for solution accuracy when WMC was set to a low level, treatment advantages relative to the control condition were found for measures of schema activation. The results indicated that the effectiveness of generative strategies among children at risk for MD was directly dependent on the level of WMC.