Abstract: Chilling is the third weather disaster following flood and typhoon in Guangdong Province. Prediction of chilling return period is of practical significance for scientific reduction and protection of disaster. Four models, including Gumbel distribution, Weibull distribution, log-normal distribution and Peasron-III distribution, were applied, based on the chilling index, to fit the probability distribution of chilling extreme calculated by chilling accumulation for 86 weather stations of Guangdong Province from 1961 to 2015 (December to the following February). The optimal models were selected to calculate the chilling extreme value of return periods. Results showed that Pearson-III distribution was the optimal model for 77 out of the 86 weather stations. The log-normal distribution was optimal for eight weather stations and Gumbel distribution was optimal for only one station. Weibull distribution was not suitable for modeling extreme value of Guangdong Province. Different return periods of 10-, 25-, 50- and 100-year were predicted by optimal distribution models respectively, with a relative error less than 6%. Chilling extreme for years presented obviously latitude distribution feature, with more in north side and less in south side, which matched the distributions of the lowest temperature, average temperature and temperature dipping scale during chilling period. Our results are useful for guiding the chilling defense for relevant industries in Guangdong Province.