The traditional probability distribution selection method is based on the observation data and plotted on the probability paper of a hypothetical distribution, which cannot conduct the direct comparison analysis due to the limitation of the probability paper. In this paper, a new generalized unified probability plot (GUPP) method is proposed. Through the Rosenblatt transform, the hypothesis distribution is transformed into a unified probability paper for visual comparison, and an efficient test method and comparison parameters are proposed based on GUPP which can be used for quick fitting and testing of large data volumes and large-scale hypothetical probability distributions. In this research, the proposed method is also used to statistically analyze and fit the real-time wind speed data collected in Xiamen from 1953 to 2015. The results show that the Pearson-Ⅲ distribution is better for annual average maximum 10-minute wind speed in Xiamen and for determining the maximum wind speed for structural design in Xiamen from different regression periods.