A new approach for tracking cell migration by performing tracking task in a blob feature space instead of the original image is proposed. The feature space is obtained through multi-scale blob detection, where the bright modes correspond to cells in the original image. Mode seeking is realized using a mean shift algorithm. Besides cell tracking, the method considers cell disappearing, appearing from the view, cell splitting and merging as well, the events that frequently happen in an image sequence. Results obtained from a real dataset indicate that the method performs well and with an acceptable speed. The method provides a useful tool for tracking cell migration automatically.