In Insurance industry, data redundancy is an extremely common problem in the population statistics. As a result a satisfactory clustering quality can rarely be obtained with the traditional clustering method. To handle this kind of problems a clustering model based on attributes reduction and SOM neural network was proposed. Using attributes reduction rules redundant information can be easily distinguished and essential attributes effectively located. And therefore the clustering quality can also be improved evidently. Experiments conducted in the H life insurance company show the method can cope with the problems mentioned above effectively.