Over a 4-year period (1982 to 1986), 91 patients with solitary or multiple metastases from colorectal cancer were stratified, based on findings at laparotomy, to one of three groups and then prospectively randomized to one of two treatment arms within each group. Group A patients had solitary resectable metastases, group B patients had multiple, resectable metastases, and group C patients had multiple, unresectable metastases. Patients were randomized to one of two treatment arms within a group: group A-arm A1: resection only, arm A2: resection and continuous hepatic artery infusion (CHAI) of fluorodeoxyuridine (FUdR); group B-arm B1: resection and CHAI, arm B2: CHAI only; group C-arm C1: CHAI, arm C2: systemic fluorouracil followed by CHAI. Median time to failure (TTF) was 31.8, 11.1, and 8.8 months for groups A, B, and C, respectively. Arm A2 had an improved TTF when compared with arm A1 (P = .03). Median survival correlated with extent of disease and was 37.3, 22.4, and 13.8 months for groups A, B, and C, respectively. Survival was not changed by treatment variation (arms) within each group. Two- and 5-year cumulative survivals for groups A, B, and C were 72.7% and 45.4%; 45.8% and 16.7%; and 31.7% and 3.2%, respectively. In patients with multiple metastases (groups B and C), those patients whose original tumor was a Dukes' B had a significantly improved TTF and survival over those patients whose tumor was a Dukes' C (P less than or equal to .02).