OBJECTIVE: To describe the main characteristics of victims, roads and vehicles involved in traffic accidents and the risk factors involved in accidents resulting in death. METHODS A non-concurrent cohort study of traffic accidents in Fortaleza, CE, Northeastern Brazil, in the period from January 2004 to December 2008. Data from the Fortaleza Traffic Accidents Information System, the Mortality Information System, the Hospital Information System and the State Traffic Department Driving Licenses and Vehicle database. Deterministic and probabilistic relationship techniques were used to integrate the databases. First, descriptive analysis of data relating to people, roads, vehicles and weather was carried out. In the investigation of risk factors for death by traffic accident, generalized linear models were used. The fit of the model was verified by likelihood ratio and ROC analysis. RESULTS There were 118,830 accidents recorded in the period. The most common types of accidents were crashes/collisions (78.1%), running over pedestrians (11.9%), colliding with a fixed obstacle (3.9%), and with motorcycles (18.1%). Deaths occurred in 1.4% of accidents. The factors that were independently associated with death by traffic accident in the final model were bicycles (OR = 21.2, 95%CI 16.1;27.8), running over pedestrians OR = 5.9 (95%CI 3.7;9.2), collision with a fixed obstacle (OR = 5.7, 95%CI 3.1;10.5) and accidents involving motorcyclists (OR = 3.5, 95%CI 2.6;4.6). The main contributing factors were a single person being involved (OR = 6.6, 95%CI 4.1;10.73), presence of unskilled drivers (OR = 4.1, 95%CI 2.9;5.5) a single vehicle (OR = 3.9, 95%CI 2,3;6,4), male (OR = 2.5, 95%CI 1.9;3.3), traffic on roads under federal jurisdiction (OR = 2.4, 95%CI 1.8;3.7), early morning hours (OR = 2.4, 95%CI 1.8;3.0), and Sundays (OR = 1.7, 95%CI 1.3;2.2), adjusted according to the log-binomial model. CONCLUSIONS Activities promoting the prevention of traffic accidents should primarily focus on accidents involving two-wheeled vehicles that most often involves a single person, unskilled, male, at nighttime, on weekends and on roads where they travel at higher speeds.