Based on the Tropical Cyclone (TC) frequency data during flood seasons in Hainan, NCEP/NCAR monthly reanalysis data and CFSv2 model historical return data, the characteristics of TC frequency and the corresponding atmospheric circulation were analyzed, moreover, the TC frequency prediction model was built based on the effective information of CFSv2 by using stepwise regression. Results show that TC frequency anomaly during flood seasons in Hainan is closely related to the large scale circulation during the corresponding period, and CFSv2 model has a great prediction skill on the key influence areas of circulation fields, including sea level pressure, geopotential height field at 500 hPa, low level wind field from the South China Sea to tropical Pacific, and zonal wind shear in the tropical Pacific. On these grounds, the correlation of TC frequency prediction model between observation and prediction of 26 years cross-validation is 0.88, and the rate of same anomalies between them reaches 88%; there are only two years that the predicted anomaly is contrary to the observed one in six years predictive testing. It can see that the prediction model has a good stability and prediction skill, which can provide effective basis for the TC frequency prediction of Hainan during flood seasons.