Working memory training is a useful tool to examine dissociations between specific working memory processes. Although current models propose a distinction between modality-specific working memory processes, to our knowledge no study has directly examined the effects of visual versus auditory working memory training. Functional magnetic resonance imaging was used to investigate whether visual working memory processes can be trained specifically and whether those effects can be separated from across-modal training effects. We found decidedly larger training gains after visual working memory training compared with auditory or no training on a visual 2-back task. These effects were accompanied by specific training-related decreases in the right middle frontal gyrus arising from visual training only. Likewise, visual and auditory training led to decreased activations in the superior portion of the right middle frontal gyrus and the right posterior parietal lobule. We infer that the combination of effects resulted from increased neural efficiency of intra-modal (visual) processes on the one hand and of across-modal (general control) processes on the other hand. Therefore, visual processes of working memory can be trained specifically, and these effects can be functionally dissociated from alterations in general control processes common to both working memory trainings.