Queue length is one of the key components in traffic monitoring and signal control at arterial intersections. For congested links, queues are difficult to measure or estimate from either loop detectors or mobile sensors, as they may exceed the region of detection. In this paper, a queue length estimation model is proposed to solve this long queue problem using short vehicle trajectories. The authors first introduce the vehicle trajectory reconstruction model to estimate the missing part in vehicle deceleration or acceleration. The long queue model is then reduced to an ordinary short queue model. A delay-based estimation method is developed to evaluate the cycle-by-cycle queue length. The proposed method is applicable to the moving queue or over-saturation condition, and performs well under low penetration rate. The model is tested in a field experiment with reasonable results.