This phase I/II study was conducted to determine the maximum recommended dose of pemetrexed when given in combination with a fixed dose of cisplatin, and the efficacy, toxicity and association of 5,10-methylenetetrahydrofolate reductase (MTHFR) variants with this pemetrexed--cisplatin combination, in patients with unresectable, advanced gastric carcinoma. Patients 18-70 years of age, with stage IV disease or post-surgery recurrence, no earlier palliative chemotherapy, 0 or 1 Eastern Cooperative Oncology Group performance status, were included. The cisplatin dose was 75 mg/m. In phase I, the initial dose of pemetrexed was 600 mg/m, escalated in 100 mg/m increments. In phase II, efficacy, including overall response rate, overall survival, as well as toxicity and MTHFR pharmacogenetics were investigated. Phase I enrolled 16 patients; 700 mg/m was defined as pemetrexed recommended dose. Thirteen serious adverse events were reported; the most common grade 3/4 toxicities were haematologic (10 of 13, 76.9%). Phase II enrolled 73 patients, 69 qualified for safety and 68 for efficacy analysis; 65 for pharmacogenomic analysis. Overall response rate was 23.5% (14.1%, 35.4%), disease control rate 55.9%, median overall survival 11.8 months (95% confidence interval, 7.2-18.5 months), progression-free survival 4.9 months (95% confidence interval, 2.8-7.1 months), and median response duration 5.4 months. Patients with MTHFR A1298C variants had median overall survival of 6.6 months, significantly shorter than patients with the wild type (median 18.5 months, P=0.001). The pemetrexed--cisplatin combination in patients with advanced gastric cancer generates modest efficacy and a manageable toxicity profile. The reduced overall survival in patients with MTHFR A1298C polymorphism variants deserves further investigation.