This paper describes the development of accident prediction models for esti
mating the safety performance of urban arterial roadways in the Greater Van
couver Regional District, British Columbia, Canada. The traffic- and road-r
elated variables that appear to underlie the occurrence of accidents are ex
amined and models that explain the occurrence of accidents as a function of
these variables are developed. To determine which variables have a signifi
cant effect on the safety of urban arterials, the study investigated a larg
e number of models with different combinations of these variables. The inve
stigation led to the conclusion that, among those variables examined, the o
nes that had a significant effect on accident occurrence were section lengt
h, traffic volume, unsignalized intersection density, driveway density, ped
estrian crosswalk density, number of traffic lanes, type of median, and typ
e of land use. The models were used to investigate the effect of the median
type on accident occurrence. The study estimated that conversion from an u
ndivided arterial to one with a raised-curb median could result, on average
, in a 10% accident reduction.