This study's purpose was to investigate effects of 3 intervention approaches for building working memory (WM) and improving word-problem solving (WPS). Children with mathematics difficulties (n = 240; 7.51 years [SD = 0.33]) were randomized to 4 conditions: a control group, general WM training with contiguous math practice, WPS intervention without WM training, and WPS intervention with domain-specific WM training. WM, WPS, and arithmetic were assessed before and 1-3 weeks after intervention; delayed WPS and arithmetic posttesting occurred 4-6 weeks later. Multilevel modeling of main effects and mediation effects were employed. Compared to control, general WM training with contiguous math practice and WPS intervention without WM training increased WM and WPS. The 3rd training condition, WPS intervention with domain-specific WM training, which minimized WM training time, improved WPS but without effects on WM. Both WPS intervention conditions outperformed general WM training on WPS. Conclusions are as follows. (a) General WM training with contiguous math practice improves WM and WPS. (b) WM training is not a substitute for WPS intervention when the goal is to strengthen WPS. (c) WPS intervention without WM training improves WM but is not a substitute for WM training when the goal is to strengthen WM. (d) For WM effects to accrue, WM training needs to occur with sufficient intensity. (e) WM plays a causal role in WPS, but not in arithmetic. Implications are drawn for research and practice, including assessing instructional supports in future research to build cognitive-academic bidirectionality. [For the grantee submission, see ED620371.]