ObjectiveBy studying the surgical outcome of deep brain stimulation (DBS) of different target nuclei for patients with refractory epilepsy, we aimed to explore a clinically feasible target nucleus selection strategy. MethodsWe selected patients with refractory epilepsy who were not eligible for resective surgery. For each patient, we performed DBS on a thalamic nucleus [anterior nucleus of the thalamus (ANT), subthalamic nucleus (STN), centromedian nucleus (CMN), or pulvinar nucleus (PN)] selected based on the location of the patient's epileptogenic zone (EZ) and the possible epileptic network involved. We monitored the clinical outcomes for at least 12 months and analyzed the clinical characteristics and seizure frequency changes to assess the postoperative efficacy of DBS on the different target nuclei. ResultsOut of the 65 included patients, 46 (70.8%) responded to DBS. Among the 65 patients, 45 underwent ANT-DBS, 29 (64.4%) responded to the treatment, and four (8.9%) of them reported being seizure-free for at least 1 year. Among the patients with temporal lobe epilepsy (TLE, n = 36) and extratemporal lobe epilepsy (ETLE, n = 9), 22 (61.1%) and 7 (77.8%) responded to the treatment, respectively. Among the 45 patients who underwent ANT-DBS, 28 (62%) had focal to bilateral tonic-clonic seizures (FBTCS). Of these 28 patients, 18 (64%) responded to the treatment. Out of the 65 included patients, 16 had EZ related to the sensorimotor cortex and underwent STN-DBS. Among them, 13 (81.3%) responded to the treatment, and two (12.5%) were seizure-free for at least 6 months. Three patients had Lennox-Gastaut syndrome (LGS)-like epilepsy and underwent CMN-DBS; all of them responded to the treatment (seizure frequency reductions: 51.6%, 79.6%, and 79.5%). Finally, one patient with bilateral occipital lobe epilepsy underwent PN-DBS, reducing the seizure frequency by 69.7%. SignificanceANT-DBS is effective for patients with TLE or ETLE. In addition, ANT-DBS is effective for patients with FBTCS. STN-DBS might be an optimal treatment for patients with motor seizures, especially when the EZ overlaps the sensorimotor cortex. CMN and PN may be considered modulating targets for patients with LGS-like epilepsy or occipital lobe epilepsy, respectively.