A computer program to detect vehicular incidents that occur on urban freeways during low-volume conditions was developed, tested, and evaluated. The type of incidents to be detected were those that involved vehicles that entered a freeway but, for some reason, did not pass through a defined study section. The algorithm can operate in real time and is based on an individual-vehicle input-output process. It was tested on traffic data from a four-lane section of freeway in Houston, Texas. The algorithm's performance was evaluated over a wide range of traffic volumes (100-1200 vehicles/h) and three different detector spacings--152.4, 304.8, and 457.2 m (500, 1000, and 1500 ft). In the 152.4-m section, the algorithm detected 65 percent (11/17) of the incidents that occurred. In the 304.8-m section, it detected 78 percent (14/18) of the incidents that occurred and, in the 457.2-m section, it detected 49 percent (17/35) of the incidents that occurred. Numerous lane changes and a bad detector at the first detection station caused the relatively poor performance of the algorithm in both the 152.4-m and the 457.2-m sections. The algorithm detected all incidents that occurred when the volume level was less than 400 vehicles/h. It detected 61 percent (37/61) of the incidents that occurred when the volume level was between 800 and 1200 vehicles/h.