Dengue is a major public health problem in tropical and subtropical areas worldwide. There is a lack of information on the risk factors for death due to severe dengue fever in developing countries, including Brazil where the state of Amazonas is located. This knowledge is important for decision making and the implementation of effective measures for patient care. This study aimed to identify factors associated with death among patients with severe dengue, in Amazonas from 2001 to 2013. We conducted a retrospective cohort study based on secondary data from the epidemiological surveillance of dengue provided by the Fundação de Vigilância em Saúde do Amazonas, FVS (Health Surveillance Foundation) of the Secretaria de Saúde do Amazonas, SUSAM (Health Secretariat of the State of Amazonas). Data on dengue cases were obtained from the SINAN (Notifiable Diseases Information System) and SIM (Mortality Information System) databases. We selected cases of severe dengue with laboratory confirmation, including dengue-related deaths of residents in the state of Amazonas from January 1, 2001, to December 31, 2013. The explanatory variables analyzed were sex, age, level of education, spontaneous hemorrhagic manifestations, plasma extravasation and platelet count. Patients who died due to severe dengue had more hematuria, gastrointestinal bleeding, and thrombocytopenia than the survivors. Considering the simultaneous effects of demographic and clinical characteristics with a multiple logistic regression model, it was observed that the factors associated with death were age >55 years (odds ratio [OR] 4.98), gastrointestinal bleeding (OR 10.26), hematuria (OR 5.07), and thrombocytopenia (OR 2.55). Gastrointestinal bleeding was the clinical sign most strongly associated with death, followed by hematuria and age >55 years. The study results showed that the best predictor of death from severe dengue is based on the characteristic of age >55 years, together with the clinical signs of gastrointestinal bleeding, hematuria, and low platelet count.