Background Docetaxel-based chemotherapy is effective in metastatic gastric and gastro-oesophageal junction adenocarcinoma, but has not yet been evaluated in the context of resectable patients. Here we report findings from the phase 2 part of the phase 2/3 FLOT4 trial, which compared histopathological regression in patients treated with a docetaxel-based triplet chemotherapy versus an anthracycline-based triplet chemotherapy before surgical resection. Methods In this randomised, open-label, phase 2/3 study, eligible participants were recruited from 28 German oncology centres. Patients with resectable gastric or gastro-oesophageal junction cancer who had clinical stage cT2 or higher, nodal positive (cN+) disease, or both were randomly assigned (1:1) to either three preoperative and three postoperative 3-week cycles of intravenous epirubicin 50 mg/m2 on day 1, intravenous cisplatin 60 mg/m2 on day 1, and either fluorouracil 200 mg/m2 as continuous intravenous infusion or capecitabine 1250 mg/m2 orally (two doses of 625 mg/m2 per day) on days 1 to 21 (ECF/ECX group) or four preoperative and four postoperative 2-week cycles of docetaxel 50 mg/m2, intravenous oxaliplatin 85 mg/m2, intravenous leucovorin 200 mg/m2, and fluorouracil 2600 mg/m2 as a 24 h infusion, all on day 1 (FLOT group). Randomisation was done centrally with an interactive web-response system based on a sequence generated with blocks (block size 2) stratified by Eastern Cooperative Oncology Group performance status, location of primary tumour, age, and nodal status. No masking was done. Central assessment of pathological regression was done according to the Becker criteria. The primary endpoint was pathological complete regression (tumour regression grade TRG1a) and was analysed in the modified intention-to-treat population, defined as all patients who were randomly assigned to treatment excluding patients who had surgery but did not provide resection specimens for central evaluation. The study (including the phase 3 part) has completed enrolment, but follow-up is ongoing and this is an interim analysis. The trial is registered with ClinicalTrials.gov, number NCT01216644. Findings Between Aug 18, 2010, and Aug 10, 2012, 300 patients (152 patients in the ECF/ECX group; 148 patients in the FLOT group) were enrolled into the phase 2 part of the study, 265 of whom (137 in the ECF/ECX group; 128 in the FLOT group) were assessable on a modified intention-to-treat basis. 119 (93%) of 128 patients in the FLOT group and 126 (92%) of 137 patients in the ECF/ECX group were given all planned preoperative cycles of treatment. FLOT was associated with significantly higher proportions of patients achieving pathological complete regression than was ECF/ECX (20 [16%; 95% CI 10–23] of 128 patients vs eight [6%; 3–11] of 137 patients; p=0·02). 44 (40%) of 111 patients in the ECF/ECX group and 30 (25%) of 119 patients in the FLOT group had at least one serious adverse event involving a perioperative medical or surgical complication. The most common non-surgical grade 3–4 adverse events were neutropenia (52 [38%] of 137 patients in the ECF/ECX group vs 67 [52%] of 128 patients in the FLOT group), leucopenia (28 [20%] vs 36 [28%]), nausea (23 [17%] vs 12 [9%]), infection (16 [12%] vs 15 [12%]), fatigue (19 [14%] vs 11 [9%]), and vomiting (13 [10%] vs four [3%]). Interpretation Perioperative FLOT was active and feasible to administer, and might represent an option for patients with locally advanced, resectable gastric or gastro-eosophageal junction adenocarcinoma. Funding None.