Background: The survival benefit and morbidity after nodal dissection for gastric cancer remains controversial. We aimed to do a single-institution randomised trial to compare D1 (ie, level 1) lymphadenectomy with that of D3 (ie, levels 1, 2, and 3) dissection for gastric cancer in terms of overall survival and disease-free survival. Methods: From Oct 7, 1993, to Aug 12, 1999, 335 patients were registered. 221 patients were eligible, 110 of whom were randomly assigned D1 surgery and 111 of whom were randomly assigned D3 surgery, both with curative intent. Three participating surgeons had done at least 25 independent D3 dissections before the start of the trial, and every procedure was verified by pathological analyses. The primary endpoints were 5-year overall survival and 5-year disease-free survival. We also analysed risk of recurrence. Main analyses were done by intention to treat. This trial is registered at the US National Institute of Health website http://www.clinicaltrials.gov/ct/show/ NCT00260884. Findings: Median follow-up for the 110 (50%) survivors was 94.5 months (range 62.9-135.1). Overall 5-year survival was significantly higher in patients assigned D3 surgery than in those assigned D1 surgery (59.5% [95% CI 50.3-68.7] vs 53.6% [44.2-63.0]; difference beteween groups 5.9% [-7.3 to 19.1], log-rank p=0.041). 215 patients who had R0 resection (ie, no microscopic evidence of residual disease) had recurrence at 5 years of 50.6% [41.1-60.2] for D1 surgery and 40.3% [30.9-49.7] for D3 surgery (difference between groups 10.3% [-3.2 to 23.7], log-rank p=0.197). Interpretation: D3 nodal dissection, compared with that of D1, offers a survival benefit for patients with gastric cancer when done by well trained, experienced surgeons.