The selection of the appropriate colorspace for tracking applications has not been an issue previously considered in the literature. Many color representations have been suggested, based on the invariance to illumination changes. Nevertheless, none of them is invariant enough to deal with general and unconstrained environments. In tracking tasks, we might prefer to represent image pixels into a colorspace where the distance between the target and background colorpoints were maximized, simplifying the task of the tracker. Based on this criterion, we propose an ?object dependent? colorspace, which is computed as a simple calibration procedure before tracking. Furthermore, this colorspace may be easily adapted at each frame. Synthetic and real experiments show how this colorspace allows for a better discrimination of the foreground and background, and permits to track in circumstances where the same tracking algorithm relying on other colorspaces would fail.