The optimal perioperative chemotherapeutic regimen for locally advanced gastric cancer remains undefined. We evaluated the efficacy and safety of perioperative and postoperative S-1 and oxaliplatin (SOX) compared with postoperative capecitabine and oxaliplatin (CapOx) in patients with locally advanced gastric cancer undergoing D2 gastrectomy.We did this open-label, phase 3, superiority and non-inferiority, randomised trial at 27 hospitals in China. We recruited antitumour treatment-naive patients aged 18 years or older with historically confirmed cT4a N+ M0 or cT4b Nany M0 gastric or gastro-oesophageal junction adenocarcinoma, with Karnofsky performance score of 70 or more. Patients undergoing D2 gastrectomy were randomly assigned (1:1:1) via an interactive web response system, stratified by participating centres and Lauren classification, to receive adjuvant CapOx (eight postoperative cycles of intravenous oxaliplatin 130 mg/m2 on day one of each 21 day cycle plus oral capecitabine 1000 mg/m2 twice a day), adjuvant SOX (eight postoperative cycles of intravenous oxaliplatin 130 mg/m2 on day one of each 21 day cycle plus oral S-1 40-60 mg twice a day), or perioperative SOX (intravenous oxaliplatin 130 mg/m2 on day one of each 21 day plus oral S-1 40-60 mg twice a day for three cycles preoperatively and five cycles postoperatively followed by three cycles of S-1 monotherapy). The primary endpoint, assessed in the modified intention-to-treat population, 3-year disease-free survival to assess the superiority of perioperative-SOX compared with adjuvant-SOX and the non-inferiority (hazard ratio non-inferiority margin of 1·33) of adjuvant-SOX compared with adjuvant-CapOx. Safety analysis were done in patients who received at least one dose of the assigned treatment. This study is registered with ClinicalTrials.gov, NCT01534546.Between Aug 15, 2012, and Feb 28, 2017, 1094 patients were screened and 1022 (93%) were included in the modified intention-to-treat population, of whom 345 (34%) patients were assigned to the adjuvant-CapOx, 340 (33%) patients to the adjuvant-SOX group, and 337 (33%) patients to the perioperative-SOX group. 3-year disease-free survival was 51·1% (95% CI 45·5-56·3) in the adjuvant-CapOx group, 56·5% (51·0-61·7) in the adjuvant-SOX group, and 59·4% (53·8-64·6) in the perioperative-SOX group. The hazard ratio (HR) was 0·77 (95% CI 0·61-0·97; Wald p=0·028) for the perioperative-SOX group compared with the adjuvant-CapOx group and 0·86 (0·68-1·07; Wald p=0·17) for the adjuvant-SOX group compared with the adjuvant-CapOx group. The most common grade 3-4 adverse events was neutropenia (32 [12%] of 258 patients in the adjuvant-CapOx group, 21 [8%] of 249 patients in the adjuvant-SOX group, and 30 [10%] of 310 patients in the perioperative-SOX group). Serious adverse events were reported in seven (3%) of 258 patients in adjuvant-CapOx group, two of which were related to treatment; eight (3%) of 249 patients in adjuvant-SOX group, two of which were related to treatment; and seven (2%) of 310 patients in perioperative-SOX group, four of which were related to treatment. No treatment-related deaths were reported.Perioperative-SOX showed a clinically meaningful improvement compared with adjuvant-CapOx in patients with locally advanced gastric cancer who had D2 gastrectomy; adjuvant-SOX was non-inferior to adjuvant-CapOx in these patients. Perioperative-SOX could be considered a new treatment option for patients with locally advanced gastric cancer.National Key Research and Development Program of China, Beijing Scholars Program 2018-2024, Peking University Clinical Scientist Program, Taiho, Sanofi-Aventis, and Hengrui Pharmaceutical.For the Chinese translation of the abstract see Supplementary Materials section.Copyright © 2021 Elsevier Ltd. All rights reserved.