BACKGROUND: To evaluate the efficacy and toxicity of a chemotherapy strategy based on histological differentiation, for patients with carcinoma of unknown primary site. PATIENTS AND METHODS: Forty-eight patients were prospectively included in the trial. Thirty patients with poorly-differentiated carcinoma or poorly-differentiated adenocarcinoma (group A) received a combination of cisplatin and etoposide. Patients with a responsive or stable disease after two cycles received the same regimen plus bleomycin, ifosfamide and G-CSF. Eighteen patients with well- or moderately-differentiated carcinoma (group B) received cisplatin, continuous infusion 5-fluorouracil (5-FU) and alpha-interferon. Treatment was maintained in case of response or stable disease for up to six cycles. RESULTS: The overall response rate (RR) for the entire group is 43% (95% confidence interval (CI): 35.9%-50.1%): seven CR and five PR in group A (RR = 40%) and six CR and two PR in group B (RR = 44%). Grade 4 leucopenia was observed in 22 (46%) patients and sepsis in 3 (6%). Median survival is 9.4 months (range 5-13.7 months) and 16.1 months (range 11.8 20.3 months), respectively. CONCLUSIONS: This chemotherapy strategy is one way to achieve high response rates, particularly for patients with well- or moderately-differentiated adenocarcinoma usually considered poorly chemosensitive.