Abstract: In the flood forecasting and warning, there are often difficulties in analyzing and calculating hydrological regime of some medium small river basins due to the shortage of hydrological information. With data acquisition of underlying surface topography, vegetation, soil and sustainable development of the data mining method, it is possible to analyze the law in the hydrology data using the unsupervised learning technique such as a cluster analysis method. Thus the parameters of its similar basins can be used in the flood forecasting of one parameter lacking basin. In this paper 118 river basins in Zhejiang Province, which have more than 20 years precipitation data, have been taken as the case studies. Using the basin length, basin width, river length, river slope, basin average slope, basin shape factor and the average maximum surface precipitation per 1 h, 3 h, 6 h and 12 h, the authors have first reduced the dimensionality using principal components analysis, and then have made the cluster analysis of the basins. The basins in Zhejiang Province are divided into 23 similar groups. On the basis of grouping， hydrological stations which have more than 20 years data of the maximum flood peak and volume are selected for comparison in order to verify whether the grouping is reasonable. The analysis results show that there is a great similarity of the maximum flood peak and volume in the similar basin groups. And the results can provide a new theory and thinking for the flood forecasting in Zhejiang Province from the point of view of statistics.