Abstract: With the purpose to improve the road traffic safety,it is important to investigate the factors which have impacts on the severity of traffic crashes.Based on the data of traffic crashes in recently years which include serious colli-sions,the types of crashes are distinguished using Poisson Model and Negative-binomial Model.These models are also used for establishing an estimation model of fatality rate for three typical crash types:rear-end crashes,lateral impact crashes and pedestrian-involved crashes.In this model,the fatality is a dependent variable and highway factors are inde-pendent variables,the numbers of persons involved in each crash is a modified variable.The parameters of this model are determined by the over-dispersion parameter and Akaike Information Criterion (AIC)method.The results indicate that the grade of highway and roadside protective facilities significantly correlate with the fatality of rear-end crashes.Weather, conditions of road surface,locations of segments or intersections,gradient of roads,and structure of pavement are all rel-evant with lateral impact crashes.The factors of conditions of road surface,grade of roads,numbers of lanes,and radius of curve are relevant with pedestrian-involved crashes.This innovated method on severity of traffic crashes proposed in this study can be further used for analyses on extremely serious traffic crashes.This study could be used as a reference for accident analysis and design of roads as well.