Based on the atmospheric weak turbulence channel model, a constellation recognition method based on fuzzy clustering and improved back propagation (BP) neural network is designed. The fuzzy C mean (FCM) algorithm is used to get the cluster center of the wireless optical multiple phase shift keying (MPSK) subcarrier signals constellation. By calculating the hard tendency of the fuzzy classification, we obtain the feature of constellation. Finally, improved BP neural network as the classifier is designed and used to accomplish the modulation recognition. When the log-amplitude fluctuation variance σχ2=0.1, the correct recognition rates of four different modulation styles are all up to 100%. With the increase of fluctuation variance, convergence of MPSK signal constellation diagram becomes worse, but the total correct recognition rate also gets to 87.5%, and the recognition rate of 16 phase shift keying (16PSK) is improved obviously.