The purposed of this research are to develop an algorithm and a prototype program to count road vehicles at night in real time using video images shooting at vehicle rear end. The proposed algorithm consists of two steps; the pre-processing step and vehicle counting step. The program counts vehicles in a predefined region or frame on the road surface. The pre-processing step allocated a vehicle in the video frame by distinguishing a vehicle from the road and forward the result to the vehicle counting step. Two approaches were used to check the existence of a vehicle. The first approach is by analyzing the rear end image of vehicles using the intensity and the red color of the vehicle rear end. The second approach is by extracting vehicle edges. The vehicle counting process used the results from the pre-processing step which was performed on two consecutive image frames. The results were then used by the counting process. In this research, video images were taken from different angles and positions. The video shooting was done on a pedestrian flyover about 7.5 metres over road surface. The resolution of video images is 320x240 pixels. The resulting vehicle counts by the program were compared to that of human counts for the two approaches. The vehicle rear end analysis approach gave the count accuracy range from 87% to 98%. The vehicle edge analysis approach gave the count accuracy range from 92% to 96%. It was found that the vehicle count accuracy depends, at least, on the shooting angles and the distance from the video camera to the detecting region, given the same quality video cameras used.