When performing upper limb reaches, the sensorimotor system can adjust to changes in target location even if the reaching limb is not visible. To accomplish this task, sensory information about the new target location and the current position of the unseen limb are used to program online corrections. Previous researchers have argued that, prior to the initiation of corrections, somatosensory information from the unseen limb must be transformed into a visual reference frame. However, most of these previous studies involved movements to visual targets. The purpose of the present study was to determine if visual sensorimotor transformations are also necessary for the online control of movements to somatosensory targets. Participants performed reaches towards somatosensory and visual targets without vision of their reaching limb. Target positions were either stationary, or perturbed before (~ 450 ms), or after movement onset (~ 100 ms or ~ 200 ms). In response to target perturbations after movement onset, participants exhibited shorter correction latencies, larger correction magnitudes, and smaller movement endpoint errors when they reached to somatosensory targets as compared to visual targets. Because reference frame transformations have been shown to increase both processing time and errors, these results indicate that hand position was not transformed into visual reference frame during online corrections for movements to somatosensory targets. These findings support the idea that different sensorimotor transformations are used for the online control of movements to somatosensory and visual targets.