Sixty-one patients with cirrhosis with varices without previous bleeding were admitted to our Department over a period of 2 yr. Fifty had alcoholic cirrhosis. Child-Turcotte-Pugh class was A in 11 patients, B in 30, and C in 20. Varices were F1 in 18, and F2-F3 in 43. During follow-up of up to 40 months, one patient was lost to follow-up and 22 patients died, seven of gastrointestinal bleeding and 15 of liver failure. The probability of death from any cause was significantly related to Child-Turcotte-Pugh class, ascites, encephalopathy, s-albumin, s-bilirubin, prothrombin index, and galactose elimination capacity. Independent prognostic variables according to Cox's model resulted s-albumin, s-bilirubin, encephalopathy, and varices. Considering only patients who died from liver failure, survival was univariately related to the same determinants, whereas the Cox's model individuated s-bilirubin, s-albumin, ascites, and galactose elimination capacity as independent prognostic indicators. Considering only patients who died from gastrointestinal bleeding no Cox model could be performed due to the small number of deaths. In patients with F1 varices, the probability of death from liver failure was 5 times higher than that from gastrointestinal bleeding; in patients with F2-F3 varices, liver failure and gastrointestinal bleeding each accounted for approximately half of the deaths. These data could become useful when programming a clinical trial of prophylaxis which considers reduction in mortality as the main end-point.