Many types of chronic pain may involve the dysregulation of complex networks in the central nervous system. The modulation of these networks may be one therapeutic route for reducing pain. Previously, we have demonstrated that patients can be trained to modulate their brain activity in a way that attenuates their experience of pain. This method involved feeding functional magnetic resonance imaging (fMRI) data back to patients in real-time. While data have shown that this real-time fMRI technique can both modulate brain activity and pain intensity, it is unknown how this training might change pain-related neuronal networks in the brain. In order to determine the impact of real-time fMRI training on pain processing, we employed a Granger Causality Mapping (GCM) algorithm. GCM allows the detection of voxels or regions of brain activity that precede or follow activity in a specific region of interest (ROI). Participants were 10 patients with chronic pain who had demonstrated an ability to modulate activity in the rostral anterior cingulate cortex (rACC). GCM analyses revealed a consistent pattern of activity across participants. At the beginning of the training program, rACC activation was preceded by prefrontal activity. At the end of the training program, however, there was no relationship between prefrontal and rACC activity. Instead, insular activity reliably preceded rACC activity. This change in brain regions showing directed influence over the rACC may reflect a shift from conscious to more automated control over brain modulation.