Neural activity in the dorsolateral prefrontal cortex (DLPFC) has been suggested to integrate information from distinct sensory areas. However, how the DLPFC interacts with the bilateral primary somatosensory cortices (SIs) in tactile-visual cross-modal working memory has not yet been established. In the present study, we applied single-pulse transcranial magnetic stimulation (sp-TMS) over the contralateral DLPFC and bilateral SIs of human participants at various time points, while they performed a tactile-visual delayed matching-to-sample task with a 2-second delay. sp-TMS over the contralateral DLPFC or the contralateral SI at either an sensory encoding stage [i.e. 100ms after the onset of a vibrotactile sample stimulus (200-ms duration)] or an early maintenance stage (i.e. 300ms after the onset), significantly impaired the accuracy of task performance; sp-TMS over the contralateral DLPFC or the ipsilateral SI at a late maintenance stage (1600ms and 1900ms) also significantly disrupted the performance. Furthermore, at 300ms after the onset of the vibrotactile sample stimulus, there was a significant correlation between the deteriorating effects of sp-TMS over the contralateral SI and the contralateral DLPFC. These results imply that the DLPFC and the bilateral SIs play causal roles at distinctive stages during cross-modal working memory, while the contralateral DLPFC communicates with the contralateral SI in the early delay, and cooperates with the ipsilateral SI in the late delay.