In this paper, the classical task of mobile target tracking using a pan-and-tilt camera is considered. The authors use recent results in motion-based visual servoing to deal with complex targets for which shape and texture are unknown. The first method consists of design- ing a control law directly from the estimated image motion. This leads to the computation of the pan-and-tilt acceleration necessary to reduce the tracking error. A second method, more efficient for target tracking, consists of retrieving the target position in the im- age from its estimated motion. This leads to classical image-based visual servoing. For both methods, experimental results obtained at video rate are presented and discussed.