Adaptive training of working memory (WM) using the Cogmed-RM intervention has recently shown some efficacy as an alternative treatment for ADHD, but this intervention may not be optimally designed. A recent component analysis of WM has suggested that maintenance in primary memory (PM) appears to be largely intact whereas recall from secondary memory (SM) appears to be deficient in ADHD relative to age-matched controls. However, extrapolating from basic research, there is reason to believe that Cogmed-RM may target the PM component more than the SM component; though training with spatial exercises may target the SM component more than training with verbal exercises. To investigate, participants diagnosed with ADHD were randomly assigned to either a verbal training condition (n = 24) or a spatial training condition (n = 23) using a randomized, controlled design, and both groups were instructed to complete at least 20 days of training. The PM and SM components of WM were assessed immediately before and after training using both verbal and spatial free recall tasks. The main findings showed that both versions of the intervention enhanced the maintenance of information in PM regardless of test modality, but not the recall of information from SM. Therefore, the component of WM that is improved by Cogmed-RM is not the same component of WM that is deficient in ADHD.