The aim of this paper was developing Safety Performance Functions (SPFs) of four different types of road intersections: unsignalized three-leg intersection, unsignalized four-leg intersection, signalized threeleg intersections, signalized four-leg intersection. The data on accidents and traffic volume of 28 intersections, which are under the jurisdiction of Friuli Venezia Giulia Strade S.p.A., were collected and analyzed. To model the empirical relation between crash frequency and traffic volume, a feed-forward artificial neural network (ANN), characterized by the hyperbolic tangent transfer function and one hidden layer, was used. SPFs of the analyzed road intersections have shown a slope reduction of the tangent line when annual average daily traffic increased on the main leg, namely, the road capacity is not able to meet the increase in the volume of traffic. As a result, this leads to a saturation that induces slowdowns with a lower probability of running into an accident.
