A two-stage differential evolution algorithm using dynamic niche radius is proposed for multimodal optimization, in which a two-stage annealing schedule based on the idea of conformational space annealing is designed to adjust the niche radius dynamically. Meanwhile, the optimization process is divided into two stages according to the annealing process. Thus, at the first stage, a differential vector limited mutation is used to generate the high-quality individuals to keep the population diversity, thereby facilitating multiple convergence. At the second stage, to enhance the convergence speed, a seed neighborhood mutation strategy is used to exploit the niche highly. Experiment results show that, the proposed algorithm can navigate from global exploration to local exploitation adaptively, which is an effective multimodal optimization algorithm.