BACKGROUND: Cancer gastrectomy seems to benefit type 2 diabetes; however, results are conflicting. In a prospective protocol, including retrospective information, the aim was assessment of changes in glucose profile in patients with both normal and deranged preoperative glucose homeostasis. METHODS: Patients (N = 164) with curative subtotal or total Roux-en-Y gastrectomy for gastric cancer (n = 92), or Roux-en-Y gastric bypass for morbid obesity (RYGB, n = 72) were preoperatively classified into diabetes (including prediabetes) and control group. Postoperative diabetes outcome was stratified as responsive or refractory, and results in controls were correspondingly defined as stable or new-onset diabetes (NOD), according to fasting blood glucose and HbA1c. Dietary intake and biochemical profile was documented. Statistical methods included analysis of variance, multivariate logistic regression, and propensity score matching according to postoperative weight loss. RESULTS: Age of cancer cases was 67.9 ± 11.5 years, 56.5 % males, initial body mass index (BMI) 24.7 ± 3.7, current BMI 22.6 ± 3.8 kg/m(2), and follow-up 102.1 ± 51.0 months, whereas in bariatric individuals age was 51.4 ± 10.1 years, 15.3 % males, initial BMI 56.7 ± 12.2, current BMI 34.8 ± 8.1 kg/m(2), and follow-up 104.1 ± 29.7 months. Refractory disease corresponded to 62.5 % (cancer) versus 23.5 % (bariatric) (P = 0.019), whereas NOD represented 69.2 versus 23.8 % respectively (P = 0.016). Weight loss (ΔBMI) was associated with diabetes response in cancer patients but not with NOD. No difference between subtotal and total gastrectomy was detected. Divergent outcomes (refractory vs. responsive) were confirmed in BMI-similar, propensity-matched cancer gastrectomy patients with preoperative diabetes, consistent with weight-dependent and -independent benefits. CONCLUSIONS: Diabetes response was confirmed, however with more refractory cases than in bariatric controls, whereas high proportions of NOD occurred. Such dichotomous pattern seems unusual albeit consistent with previous studies.