A total of 163 patients with advanced cancer at an Australian teaching hospital were interviewed to investigate whether or not emotional support was predictive of survival duration. Survival was analysed using the Kaplan–Meier product-limit estimate and multivariable Cox proportional hazards regression from entry to the study in 1996 to date of death or 31 March 2003, whichever came first. The number of confidants with whom feelings were being shared at the time of study entry was predictive of survival duration. The regression analysis indicated that compared with patients reporting two or three confidants, the relative risk of a shorter survival was 0.44 (95% CI 0.25 to 0.79) for those with no or one confidant and 0.60 (0.40, 0.89) for those with four or more confidants.