Under the impact of climate change, extreme events occur frequently and the speed of alternation of drought and flood gets faster. Based on the daily rainfall data of 42 meteorological stations for the years 1963-2013, Haihe River basin in China was divided into four sub-regions (region A, region B, region C, region D) by means of cluster analysis of the meteorological stations using the self-organizing map (SOM) neural network method. The drought index of the number of consecutive days without rainfall was used to identify the drought magnitudes and the first rainfall intensity and the maximum daily rainfall after a drought period was analyzed with the Pearson-III frequency curve, then the relationship between rainfall intensity and different drought magnitudes was observed. The results indicated that: (1) Annual rainfall reduced significantly and the drought number increased in the Haihe River basin, meanwhile the occurrence frequency of different drought level showed an overall increasing trend. This region was dominated by light drought, with the probability as 76%, while the probability of moderate drought was about 20%. The spatial difference was significant. The occurrence frequencies of moderate drought and severe drought were higher in region A. The frequency of extraordinary drought was the most in region B as 0.35%. (2) As the drought level increased, the probability of the first effective rainfall reaching heavy rain decreased in region A, region B and region C, while the probability increased in region D. With the increase of the drought number, rainfall frequency decreased and rainfall intensity enhanced obviously.