The measurement of vehicle turn ratio and based on that a prediction of the probability of a car to make a turn at an intersection are useful information for advanced traffic prediction and control. Conventional vision-based vehicle detection procedures that merely count the number of passing vehicles are not sufficient for turn ratio measurement, because they cannot tell where the vehicles come from at an intersection. This paper presents an image-based system for real-time turn ratio measurement. A method is proposed to estimate automatically the turn ratio of an intersection by using the techniques of detection window and optical flow calculation of motion vectors. A stand-alone image measurement system has been installed at a busy road junction in Hsinchu Science Park. Experimental results reveal that the average measurement error is within 2%.