For patients with gastric cancer deemed curable the only treatment option is surgery, but there is disagreement about whether accompanying lymph-node dissection should be limited to the perigastric nodes (D1) or should extend to regional lymph nodes outside the perigastric area (D2). We carried out a multicentre randomised comparison of D1 and D2 dissection. 1078 patients were randomised (539 to each group). 26 allocated D1 and 56 allocated D2 were found not to satisfy eligibility criteria (histologically confirmed adenocarcinoma of the stomach without clinical evidence of distant metastasis). Each of the remainder was attended by one of eleven supervising surgeons who decided whether curative resection was possible and, if so, assisted with the allocated procedure. Among the 711 patients (380 D1, 331 D2) judged to have curable lesions, D2 patients had a higher operative mortality rate than D1 patients (10 vs 4%, p = 0.004) and experienced more complications (43 vs 25%, p < 0.001). They also needed longer postoperative hospital stays (median 25 [range 7-277] vs 18 [7-143] days, p < 0.001). Morbidity and mortality differences persisted in almost all subgroup analyses. While we await survival results, D2 dissection should not be used as standard treatment for western patients.