Transcatheter arterial chemoembolization (TACE) improves the treatment of hepatocellular carcinoma (HCC) by causing tumour necrosis and shrinkage. Fifty-two patients with resectable large HCC (defined as a maximal tumour diameter of 10 cm or more) were prospectively randomized into two groups: group 1 comprised 24 patients who had 1–5 sessions of TACE before operation; group 2 consisted of the other 28 patients, on whom surgery was performed without delay. Tumour volume was reduced to a mean(s.d.) of 42·8(15·3) per cent in 16 patients in group 1, but remained unchanged in four and increased in size in a further four. Patients in group 1 had a slightly longer operating time (5·5 versus 4·6 h, P = 0·09), a higher rate of concomitant resection of adjacent organs (58 versus 25 percent, P = 0·03) and a higher rate of histological invasion to these organs (33 versus 4 per cent, P = 0·01). No difference was found between the two groups in operative blood loss, operative morbidity and mortality rates, and pathological staging. The disease-free survival rate in the two groups was similar, but the incidence of extrahepatic cancer recurrence was higher in group 1 (57 versus 23 per cent, P = 0·03). The actuarial survival rate was also significantly worse in group 1 when determined from the time of detection of the tumour (P = 0·03) or from operation (P = 0·01). It is concluded that preoperative TACE for resectable large HCC should be avoided because it does not provide complete necrosis in large tumours and results in delayed surgery and difficulty in the treatment of recurrent lesions, without any benefit.