Background: There is no standard treatment for unresectable hepatocellular carcinoma. Arterial embolisation is widely used, but evidence of survival benefits is lacking. Methods: We did a randomised controlled trial in patients with unresectable hepatocellular carcinoma not suitable for curative treatment, of Child-Pugh class A or B and Okuda stage I or II, to assess the survival benefits of regularly repeated arterial embolisation (gelatin sponge) or chemoembolisation (gelatin sponge plus doxorubicin) compared with conservative treatment. 903 patients were assessed, and 112 (12%) patients were finally included in the study. The primary endpoint was survival. Analyses were by intention to treat. Findings: The trial was stopped when the ninth sequential inspection showed that chemoembolisation had survival benefits compared with conservative treatment (hazard ratio of death 0.47 [95% CI 0.25-0.91], p=0.025). 25 of 37 patients assigned embolisation, 21 of 40 assigned chemoembolisation, and 25 of 35 assigned conservative treatment died. Survival probabilities at 1 year and 2 years were 75% and 50% for embolisation; 82% and 63% for chemoembolisation, and 63% and 27% for control (chemoembolisation vs control p=0.009). Chemoembolisation induced objective responses sustained for at least 6 months in 35% (14)of cases, and was associated with a significantly lower rate of portal-vein invasion than conservative treatment. Treatment allocation was the only variable independently related to survival (odds ratio 0.45 [95% CI 0.25-0.81], p=0.02). Interpretation: Chemoembolisation improved survival of stringently selected patients with unresectable hepatocellular carcinoma.