Globus pallidus internus (GPi) deep brain stimulation (DBS) and subthalamic nucleus (STN) DBS have risen as promising treatment options for patients with cervical dystonia (CD). To map the prime stimulation sites and develop a neural network, 38 CD patients undergoing GPi-DBS and STN-DBS were employed, with data from 25 patients used for network training and that of 13 patients used for external validation. CD improvements were determined by the Toronto Western Spasmodic Torticollis Rating Scale (TWSTRS), efficacy of stimulation targets were identified by a probabilistic stimulation map, structural and functional connectivity analyses were executed using a normative diffusion MRI connectome (Human Connectome Project) and normative resting-state functional MRI connectome (Brain Genomics Superstruct Project), and model predictive values were assessed by the leave-one-patient-out cross-validation (LOOCV) method. The posterior ventral medial GPi, and part of the posterior lateral GPi, were identified as optimal targets for GPi-DBS, with suboptimal benefit from stimulation in the globus pallidus externus (GPe) and lateral GPi. For STN-DBS, the prime target resided in the dorsolateral STN and worsened symptoms were associated with stimulation of the lateral and ventral medial sides of the STN. In a combined dataset, beneficial fiber tracts were found projecting to the primary motor cortex, premotor cortex, dorsolateral prefrontal cortex, and cerebellum, while negatively associated ones projected to the somatosensory cortex and visual cortex. Structural and functional models of the combined datasets were able to identify improvements and deterioration of fiber tracts or brain regions, but models of solely GPi-DBS or STN-DBS failed to do so. There was discussion about similar optimal stimulation sites for STN-DBS in CD and Parkinson’s disease patients, but refinement of electrode implantation sites for each requires further investigation as they do not entirely overlap. Additionally, while STN-DBS is currently an alternative for GPi-DBS, the findings may suggest that STN‐DBS has a stronger influence on the premotor cortices than GPi‐DBS.