In oncology, combating the spread of tumor cells is a clinical need which currently remains unsatisfied. Identifying anti-migratory compounds usually requires in vitro screening of a large number of molecules. Efficient and realistic (i.e., preferably 3D) in vitro tests are thus required in order to quantify the anti-migratory effects of anti-cancer drugs. To remain compatible with high-throughput screening, we focus on assays where unlabeled cells are migrating in 3D transparent gels and are observed under time-lapse 3D phase-contrast microscopy. In this context, we present a method for automatically tracking cells that combines a template matching preprocessing step with a mean-shift process. The preprocessing step consists in performing a correlation of a cell template with each observed volume in order to provide a phase-contrast artifact-free volume where the cells appear as correlation peaks surrounded by smooth gradients. This transformation enables the cells to be efficiently tracked by a mean-shift process. Robustness and efficiency of this approach are qualitatively and quantitatively shown in various experiments. Finally, we successfully applied our method to the quantitative characterization of the anti-migratory impact of cytochalasin-D on cancer cells. In conclusion, our method can efficiently be used for drug screening aiming to evidence drug-induced effects on cell migration in 3D transparent environments, such as matrix gels.