Objective: Conduct a survival analysis of elderly patients hospitalized in an intensive care unit (ICU), identifying the predictors of mortality among this age group. Methods: A retrospective cohort study was performed with data from the medical records of 457 elderly patients hospitalized in an ICU located in the city of Natal in Brazil. Survival functions were estimated using the Kaplan-Meier estimator, and the Log-rank test was used for comparisons. In addition, a multiple Cox proportional hazards model was constructed to identify the independent effects of the predictors of survival. Results: It was found that the survival of elderly ICU patients declined due to factors such as increased hospitalization time, advancing years, unmarried (including common-lawmarriage) status, the presence of shock, pneumonia, septicaemia, fractures, a reduced state of consciousness, hospitalization for clinical reasons, being bedridden prior to hospitalization, fever, bradycardia, hypotension, cardiac arrest and the need for mechanical ventilation. The multiple Cox proportional hazards model revealed that variables such as shock, longevity, bradycardia, fractures, fever, hospitalization in the public healthcare system and admission for clinical reasons remained significant as predictors of reduced survival in intensive care units. Conclusions: The survival rates of elderly persons in an ICU in the city of Natal in Brazil were affected by demographic and clinical predictors, and those related to the type of hospitalization and the health care network. This shows that any initiative aimed at increasing the survival of elderly ICU patients must look at individual and social issues and factors related to the health care network.