Between February 1978 and January 1984, 222 eligible patients were randomised in a multicentre trial of preoperative intra-arterial chemotherapy in the treatment of oral cavity and oropharynx carcinoma. Patients were randomised between either surgery or preoperative chemotherapy. This latter group received vincristine and bleomycin for 12 days. Patients were stratified according to the primary site: floor of the mouth (FM) versus posterior oral cavity or oropharynx (POC) and institution. The FM group received postoperative radiotherapy depending upon quality of the margins and lymph-node pathological involvement, when it was systematically applied in the POC group. Tumour regression after chemotherapy either complete (CR) or partial (PR greater than 50%) was observed in 48% in the FM group and 41% in the POC group, and lymph-node regression (CR + PR) was respectively 15% and 23%. Some discrepancies appeared between clinical regression and pathological response, and the number of cases without histological response was clearly higher than the number of cases without clinical response. The overall survival showed a statistically significant difference (P = 0.048) between FM and POC groups. In the FM group, median survival in the chemotherapy arm was estimated at 7 years compared with 3 years in the surgery arm. In the POC group, median survival was estimated at 3 years in both treatment arms. Chemotherapy lowered the uncontrolled disease and local recurrence in the FM group. These differences do not exist in the POC group, which may be due to the systematically postoperative radiotherapy.