This paper focuses on the risk decision-making problem in customs targeting, whose major responsibility is to inspect the smuggling goods in import/export declarations. In this problem, the estimated smuggling probabilities of import/export goods, which can be obtained by applying statistical analysis to observations (samples), are needed for accurate inspection decision. A critical presumption for statistical analysis is that the samples are homogeneous or subject to certain distributions. Therefore, clustering techniques are usually employed for preprocessing the samples. However, severe heterogeneity and abnormality exist among the large amount of samples and thus hinder the performance of conventional clustering methods for preprocessing. To deal with this problem, a dynamic K-means clustering method is developed in this paper. Through optimizing the validity function that indicates the goodness of clustering result, the entire samples are iteratively divided into a number of clusters. Based on the dynamic K-means clustering method and logistic regression, a risk decision-making approach is proposed and applied to China's customs targeting. The empirical results show that the proposed approach improves the accuracy and decreases the risk of inspection decision.