Pedestrian Dead Reckoning (PDR) is a core component in pedestrian navigation. Usually, PDR algorithms use the current position and movement information to figure out position in the future in order to accomplish the navigation task. Step detection, as a basic portion of PDR, is significant for the implementation of Pedestrian Navigation. In this paper, a step detection algorithm is designed based on the existing research in the relative area. To improve accuracy, the algorithm involves a Fast Fourier Transformation (FFT) for optimizing. At last, an experiment is conducted for this algorithm, and the error rate of step detection is less than 1%.