Real-world datasets are not idealized, which may exhibit irregular and nonlinear patterns in their spatial distribution, and the boundaries between different categories of data points may overlap or even be ambiguous. However, most clustering algorithms are based on Euclidean distance. For nonlinear distributed data, such as interleaved data and high-dimensional data in Euclidean space, similarity cannot be calculated efficiently. In view of KNN’s good processing ability for local information and the advantage of granular computing to deal with uncertain information, a hierarchical clustering algorithm based on KNN and granular computing is proposed in this paper. In this method, the neighbor information of the data points is judged by constructing the similarity matrix of hesitant difference granularity, and the similarity between data is redefined by sharing the neighbor. Finally, through comparative analysis, it is proved that the method can identify the interleaved datasets in Euclidean space more accurately and has better robustness.