Objectives: to determine the nutritional evaluation method that best predicts mortality in 90 days of patients submitted to gastrectomy for gastric cancer. Methods: we conducted a prospective study with 44 patients with gastric cancer, stages II to IIIa, of whom nine were submitted to partial gastrectomy, 34 to total gastrectomy, and one to esophago-gastrectomy. All patients were nutritionally evaluated through the same protocol, up to 72h after hospital admission. The parameters used were Patient-Generated Subjective Global Assessment (PGSGA), classical anthropometry, current weight and height, percentage of weight loss (%WL) and body mass index (BMI). We also measured the thickness of the thumb adductor muscle (TAM) in both hands, dominant hand (TAMD) and non-dominant hand (TAMND), as well as the calculated the prognostic nutritional index (PNI). The laboratory profile included serum levels of albumin, erythrocytes, hemoglobin, hematocrit, leukocytes, and total lymphocytes count (TLC). Results: of the 44 patients studied, 29 (66%) were malnourished by the subjective method, 15 being grade A, 18 grade B and 11 grade C. Cases with PGSGA grade B and TAMD 10.2±2.9 mm were significantly associated with higher mortality. The ROC curves (95% confidence interval) of both PGSGA and TAMD thickness reliably predicted mortality at 30 and 90 days. No laboratory method allowed predicting mortality at 90 days. Conclusion: PGSGA and the TAMD thickness can be used as preoperative parameters for risk of death in patients undergoing gastrectomy for gastric cancer.