This paper is concerned with the multi-target localization based on compressive sensing (CS). If orthogonal matching pursuit (OMP) is applied as the signal reconstruction algorithm to locate targets, there will always be a situation that only part of targets are located successfully. To solve this problem, divided localization of multiple targets is proposed. In data preparation stage, the results of single and multiple localization are calculated under two kinds of partitioning granularity. Then, in divided localization stage, according to the data from data preparation stage, part of targets can be predetermined by finding the clusters. Also, the remaining targets can be located in the possible falling area. Finally, simulation results are provided to demonstrate that, compared with other schemes, the proposed algorithm obviously improves the positioning accuracy and the anti-noise ability.