Abstract: Expressways are the backbones of the urban traffic network.The safety situation on expressways can di-rectly affect the efficiency of traffic.50 069 accidents occurred in the year 2011-2013 is collected by central video monitor system in a three-year period.Based on traffic flow data from dual-loop detectors and accidents data,a Bayesian spatial model is developed to analyze four expressways in the City of Shanghai:North-South elevated road,Yan′an elevated road,Inner Ring,and Middle Ring.In order to reveal the potential spatial correlation among road segments,an adjacent-correlation spatial model is developed by adding a spatial random-effect factor to the negative binomial model.Further,a spatial correlation model of distance is developed by improving the proximity matrix of the factor.The Bayesian method is applied to estimate the parameters and compare the regression results of these three models.The results show that the spatial correlation model of distance is superior in terms of goodness-of-fit.The deviance information criterion (DIC)of the spatial correlation model of distance is 6.04% lower than which in the negative binomial model.It indicates that the distribution of accidents on expressway segments is spatial correlated and is related to the distance between segments.An analysis on road geometric features shows that the risk of accident tends to be higher with a larger number of ramps with-in 1 km before or after the midpoint of segments;the cumulative turning angle per kilometer is larger as well.Weaving segments between on-ramps and off-ramps have a higher risk of accident.The volume of average daily traffic shows a sig-nificant positive correlation with accidents.Compared with the other two expressways,the risk of accident on Inner Ring is higher;while Middle Ring is relatively lower.