In this paper, a new method for the detection of queue lengths at traffic signals is described. Based on conventional probe vehicle data and implementing an extremely flexible data fusion approach for the integration of nearly arbitrary additional traffic information, it provides an efficient way to get high-quality estimates for the traffic states at traffic signals. A systematic evaluation based on extensive simulations addresses several issues concerning quality and demonstrates both the great potentials as well as the limitations of the new approach. In conclusion, it becomes apparent that high accuracies can be achieved even with today data.