We examined the effectiveness of (a) a working memory (WM) training program and (b) a combination program involving both WM training and direct instruction for students with geometry difficulties (GD). Four students with GD participated. A multiple-baseline design across participants was employed. During the Phase 1, students received six sessions of computerized WM training (Jaeggi et al., 2011); during Phase 2, they received six sessions of computerized WM training and six sessions of human-delivered direct instruction with Concrete--Representational--Abstract (CRA) representation sequence on triangle congruence. All four participants remarkably improved their WM, and enhanced their performance on the spatial rotation and general geometric problem-solving. However, the participants' performance on specific triangle-congruence tasks did not improve until direct instruction with CRA sequence was introduced. Practical recommendations were discussed. Practical recommendations of providing cognitive remediation combined with academic interventions were emphasized.