To segment targets under complicated background from infrared images of natural scenes, an infrared image segmentation algorithm based on mathematical morphology was presented. First, the noise and small interferential regions are removed with morphological filters. Second, a multiscale algorithm for computing morphological gradient images is proposed and the gradients are derived. Third, the images are segmented by improved watersheds and the problem of over-segmentation regions is settled well by an effective region merging method. The experiment results show that the approach performs well in target segmentation in infrared images.