During 25 years (1960 to 1984), 657 patients (aged 22 to 91, mean 66 years) were operated on for carcinomas (squamous cell, n = 230; adenocarcinoma, n = 399; anaplastic, n = 28) of the thoracic esophagus (n = 347) or gastric cardia (n = 310). Esophagogastrectomy was accomplished in 514 patients, of whom 94% (n = 481) had an "inkwell" esophagogastrostomy performed. The hospital mortality rate (less than or equal to 30 days) was 19% and the 5-year cumulative survival rate was 9% +/- 1% (standard error). A Cox regression analysis enabled a detailed risk stratification of the patients. T, N, and M class and age were the strongest predictor variables. The general status of the patients, including pulmonary disease, also had a strong prognostic influence. Eight risk groups were identified having 1-year and 5-year survival rates of 71%/41% (n = 35), 69%/24% (n = 80), 47%/11% (n = 125), 30%/6% (n = 139), 12%/0% (n = 105), 6%/0% (n = 71), 2%/0% (n = 57), and 0%/0% (n = 45) (p less than 0.0001). Hospital mortality (from 6% to 42%) and complication rates increased significantly from the low-risk to high-risk groups. Comparisons with survival rates of background populations matched to each of the first four risk groups indicated that the benefit of surgical treatment may be underestimated if only patient survivals are judged. Use of modern endoscopic and noninvasive tests may yield a reliable TNM classification without surgical exploration. Given the extremely poor prognostic outlook together with high hospital mortality and complication rates of the four last risk groups, an aggressive surgical approach with resection whenever possible can no longer be regarded rational. Selection for surgical treatment should be based on a detailed risk estimation that takes into account both TNM classification and general patient status.