From 1960 to 1992 a total of 1718 patients with liver metastases from colorectal carcinoma were recorded. Of these patients, 469 (27.3%) underwent hepatic resection, which was performed with curative intent in 434 patients (25.3%). Operative mortality in this group was 4.4%, being 1.8% (2 of 114) during the last 3 years. Significant morbidity was observed in 16% of patients with a decrease to 5% (6 of 112) for the last 3 years. A 99.8% follow-up until November 1, 1993 was achieved. Excluding operative mortality, there are 350 patients with "potentially curative" resection and 65 corresponding patients with minimal macroscopic (n = 19) or microscopic (n = 46) residual disease. The latter group demonstrated a poor prognosis, with median and maximum survival times of 14.4 and 56.0 months, respectively. Among the 350 patients having potentially curative resection, the actuarial 5-, 10-, and 20-year survivals were 39.3%, 23.6%, and 17.7%, respectively. Tumor-free survival was 33.6% at 5 years. In the univariate analysis, the following factors were associated with decreased crude survival: presence and extent of mesenteric lymph node involvement (p = 0.0001); grade III/IV primary tumor (p = 0.013); synchronous diagnosis of metastases (p = 0.014); satellite metastases (p = 0.00001); metastasis diameter of > 5 cm (p = 0.003); preoperative carcinoembryonic antigen (CEA) elevation (p = 0.03); limited resection margins (p = 0.009); extrahepatic disease (p = 0.009); and nonanatomic procedures (p = 0.008). With respect to disease-free survival, extrahepatic disease (p = 0.09) failed to achieve statistical significance, whereas patients with primary tumors in the colon did significantly better than those with rectal cancer (p = 0.04). The presence of five or more independent metastases adversely affected resectability (p < 0.05). However, once a radical excision of all detectable disease was achieved, no significant predictive value of an increasing number of metastases (1-3 versus > or = 4) on either overall (p = 0.40) or disease-free (p = 0.64) survival was found. Using Cox's multivariate regression analysis, the presence of satellite metastases, primary tumor grade, the time of metastasis diagnosis, diameter of the largest metastasis, anatomic versus nonanatomic approach, year of resection, and mesenteric lymph node involvement each independently affected both crude and tumor-free survival.